Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/ipc/sem.c | |
3 | * Copyright (C) 1992 Krishna Balasubramanian | |
4 | * Copyright (C) 1995 Eric Schenk, Bruno Haible | |
5 | * | |
1da177e4 LT |
6 | * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com> |
7 | * | |
8 | * SMP-threaded, sysctl's added | |
624dffcb | 9 | * (c) 1999 Manfred Spraul <manfred@colorfullife.com> |
1da177e4 | 10 | * Enforced range limit on SEM_UNDO |
046c6884 | 11 | * (c) 2001 Red Hat Inc |
1da177e4 LT |
12 | * Lockless wakeup |
13 | * (c) 2003 Manfred Spraul <manfred@colorfullife.com> | |
c5cf6359 MS |
14 | * Further wakeup optimizations, documentation |
15 | * (c) 2010 Manfred Spraul <manfred@colorfullife.com> | |
073115d6 SG |
16 | * |
17 | * support for audit of ipc object properties and permission changes | |
18 | * Dustin Kirkland <dustin.kirkland@us.ibm.com> | |
e3893534 KK |
19 | * |
20 | * namespaces support | |
21 | * OpenVZ, SWsoft Inc. | |
22 | * Pavel Emelianov <xemul@openvz.org> | |
c5cf6359 MS |
23 | * |
24 | * Implementation notes: (May 2010) | |
25 | * This file implements System V semaphores. | |
26 | * | |
27 | * User space visible behavior: | |
28 | * - FIFO ordering for semop() operations (just FIFO, not starvation | |
29 | * protection) | |
30 | * - multiple semaphore operations that alter the same semaphore in | |
31 | * one semop() are handled. | |
32 | * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and | |
33 | * SETALL calls. | |
34 | * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO. | |
35 | * - undo adjustments at process exit are limited to 0..SEMVMX. | |
36 | * - namespace are supported. | |
37 | * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing | |
38 | * to /proc/sys/kernel/sem. | |
39 | * - statistics about the usage are reported in /proc/sysvipc/sem. | |
40 | * | |
41 | * Internals: | |
42 | * - scalability: | |
43 | * - all global variables are read-mostly. | |
44 | * - semop() calls and semctl(RMID) are synchronized by RCU. | |
45 | * - most operations do write operations (actually: spin_lock calls) to | |
46 | * the per-semaphore array structure. | |
47 | * Thus: Perfect SMP scaling between independent semaphore arrays. | |
48 | * If multiple semaphores in one array are used, then cache line | |
49 | * trashing on the semaphore array spinlock will limit the scaling. | |
50 | * - semncnt and semzcnt are calculated on demand in count_semncnt() and | |
51 | * count_semzcnt() | |
52 | * - the task that performs a successful semop() scans the list of all | |
53 | * sleeping tasks and completes any pending operations that can be fulfilled. | |
54 | * Semaphores are actively given to waiting tasks (necessary for FIFO). | |
55 | * (see update_queue()) | |
56 | * - To improve the scalability, the actual wake-up calls are performed after | |
57 | * dropping all locks. (see wake_up_sem_queue_prepare(), | |
58 | * wake_up_sem_queue_do()) | |
59 | * - All work is done by the waker, the woken up task does not have to do | |
60 | * anything - not even acquiring a lock or dropping a refcount. | |
61 | * - A woken up task may not even touch the semaphore array anymore, it may | |
62 | * have been destroyed already by a semctl(RMID). | |
63 | * - The synchronizations between wake-ups due to a timeout/signal and a | |
64 | * wake-up due to a completed semaphore operation is achieved by using an | |
65 | * intermediate state (IN_WAKEUP). | |
66 | * - UNDO values are stored in an array (one per process and per | |
67 | * semaphore array, lazily allocated). For backwards compatibility, multiple | |
68 | * modes for the UNDO variables are supported (per process, per thread) | |
69 | * (see copy_semundo, CLONE_SYSVSEM) | |
70 | * - There are two lists of the pending operations: a per-array list | |
71 | * and per-semaphore list (stored in the array). This allows to achieve FIFO | |
72 | * ordering without always scanning all pending operations. | |
73 | * The worst-case behavior is nevertheless O(N^2) for N wakeups. | |
1da177e4 LT |
74 | */ |
75 | ||
1da177e4 LT |
76 | #include <linux/slab.h> |
77 | #include <linux/spinlock.h> | |
78 | #include <linux/init.h> | |
79 | #include <linux/proc_fs.h> | |
80 | #include <linux/time.h> | |
1da177e4 LT |
81 | #include <linux/security.h> |
82 | #include <linux/syscalls.h> | |
83 | #include <linux/audit.h> | |
c59ede7b | 84 | #include <linux/capability.h> |
19b4946c | 85 | #include <linux/seq_file.h> |
3e148c79 | 86 | #include <linux/rwsem.h> |
e3893534 | 87 | #include <linux/nsproxy.h> |
ae5e1b22 | 88 | #include <linux/ipc_namespace.h> |
5f921ae9 | 89 | |
1da177e4 LT |
90 | #include <asm/uaccess.h> |
91 | #include "util.h" | |
92 | ||
e57940d7 MS |
93 | /* One semaphore structure for each semaphore in the system. */ |
94 | struct sem { | |
95 | int semval; /* current value */ | |
96 | int sempid; /* pid of last operation */ | |
6062a8dc | 97 | spinlock_t lock; /* spinlock for fine-grained semtimedop */ |
ab63bc97 MS |
98 | struct list_head pending_alter; /* pending single-sop operations */ |
99 | /* that alter the semaphore */ | |
100 | struct list_head pending_const; /* pending single-sop operations */ | |
101 | /* that do not alter the semaphore*/ | |
bf6830ad | 102 | time_t sem_otime; /* candidate for sem_otime */ |
0824e44c | 103 | } ____cacheline_aligned_in_smp; |
e57940d7 MS |
104 | |
105 | /* One queue for each sleeping process in the system. */ | |
106 | struct sem_queue { | |
e57940d7 MS |
107 | struct list_head list; /* queue of pending operations */ |
108 | struct task_struct *sleeper; /* this process */ | |
109 | struct sem_undo *undo; /* undo structure */ | |
110 | int pid; /* process id of requesting process */ | |
111 | int status; /* completion status of operation */ | |
112 | struct sembuf *sops; /* array of pending operations */ | |
113 | int nsops; /* number of operations */ | |
114 | int alter; /* does *sops alter the array? */ | |
115 | }; | |
116 | ||
117 | /* Each task has a list of undo requests. They are executed automatically | |
118 | * when the process exits. | |
119 | */ | |
120 | struct sem_undo { | |
121 | struct list_head list_proc; /* per-process list: * | |
122 | * all undos from one process | |
123 | * rcu protected */ | |
124 | struct rcu_head rcu; /* rcu struct for sem_undo */ | |
125 | struct sem_undo_list *ulp; /* back ptr to sem_undo_list */ | |
126 | struct list_head list_id; /* per semaphore array list: | |
127 | * all undos for one array */ | |
128 | int semid; /* semaphore set identifier */ | |
129 | short *semadj; /* array of adjustments */ | |
130 | /* one per semaphore */ | |
131 | }; | |
132 | ||
133 | /* sem_undo_list controls shared access to the list of sem_undo structures | |
134 | * that may be shared among all a CLONE_SYSVSEM task group. | |
135 | */ | |
136 | struct sem_undo_list { | |
137 | atomic_t refcnt; | |
138 | spinlock_t lock; | |
139 | struct list_head list_proc; | |
140 | }; | |
141 | ||
142 | ||
ed2ddbf8 | 143 | #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS]) |
e3893534 | 144 | |
1b531f21 | 145 | #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid) |
1da177e4 | 146 | |
7748dbfa | 147 | static int newary(struct ipc_namespace *, struct ipc_params *); |
01b8b07a | 148 | static void freeary(struct ipc_namespace *, struct kern_ipc_perm *); |
1da177e4 | 149 | #ifdef CONFIG_PROC_FS |
19b4946c | 150 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it); |
1da177e4 LT |
151 | #endif |
152 | ||
153 | #define SEMMSL_FAST 256 /* 512 bytes on stack */ | |
154 | #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */ | |
155 | ||
156 | /* | |
b56e88e2 | 157 | * Locking: |
1da177e4 | 158 | * sem_undo.id_next, |
b56e88e2 | 159 | * sem_array.complex_count, |
ab63bc97 | 160 | * sem_array.pending{_alter,_cont}, |
b56e88e2 | 161 | * sem_array.sem_undo: global sem_lock() for read/write |
1da177e4 LT |
162 | * sem_undo.proc_next: only "current" is allowed to read/write that field. |
163 | * | |
b56e88e2 MS |
164 | * sem_array.sem_base[i].pending_{const,alter}: |
165 | * global or semaphore sem_lock() for read/write | |
1da177e4 LT |
166 | */ |
167 | ||
e3893534 KK |
168 | #define sc_semmsl sem_ctls[0] |
169 | #define sc_semmns sem_ctls[1] | |
170 | #define sc_semopm sem_ctls[2] | |
171 | #define sc_semmni sem_ctls[3] | |
172 | ||
ed2ddbf8 | 173 | void sem_init_ns(struct ipc_namespace *ns) |
e3893534 | 174 | { |
e3893534 KK |
175 | ns->sc_semmsl = SEMMSL; |
176 | ns->sc_semmns = SEMMNS; | |
177 | ns->sc_semopm = SEMOPM; | |
178 | ns->sc_semmni = SEMMNI; | |
179 | ns->used_sems = 0; | |
ed2ddbf8 | 180 | ipc_init_ids(&ns->ids[IPC_SEM_IDS]); |
e3893534 KK |
181 | } |
182 | ||
ae5e1b22 | 183 | #ifdef CONFIG_IPC_NS |
e3893534 KK |
184 | void sem_exit_ns(struct ipc_namespace *ns) |
185 | { | |
01b8b07a | 186 | free_ipcs(ns, &sem_ids(ns), freeary); |
7d6feeb2 | 187 | idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr); |
e3893534 | 188 | } |
ae5e1b22 | 189 | #endif |
1da177e4 LT |
190 | |
191 | void __init sem_init (void) | |
192 | { | |
ed2ddbf8 | 193 | sem_init_ns(&init_ipc_ns); |
19b4946c MW |
194 | ipc_init_proc_interface("sysvipc/sem", |
195 | " key semid perms nsems uid gid cuid cgid otime ctime\n", | |
e3893534 | 196 | IPC_SEM_IDS, sysvipc_sem_proc_show); |
1da177e4 LT |
197 | } |
198 | ||
e5639c52 MS |
199 | /** |
200 | * unmerge_queues - unmerge queues, if possible. | |
201 | * @sma: semaphore array | |
202 | * | |
203 | * The function unmerges the wait queues if complex_count is 0. | |
204 | * It must be called prior to dropping the global semaphore array lock. | |
205 | */ | |
206 | static void unmerge_queues(struct sem_array *sma) | |
207 | { | |
208 | struct sem_queue *q, *tq; | |
209 | ||
210 | /* complex operations still around? */ | |
211 | if (sma->complex_count) | |
212 | return; | |
213 | /* | |
214 | * We will switch back to simple mode. | |
215 | * Move all pending operation back into the per-semaphore | |
216 | * queues. | |
217 | */ | |
218 | list_for_each_entry_safe(q, tq, &sma->pending_alter, list) { | |
219 | struct sem *curr; | |
220 | curr = &sma->sem_base[q->sops[0].sem_num]; | |
221 | ||
222 | list_add_tail(&q->list, &curr->pending_alter); | |
223 | } | |
224 | INIT_LIST_HEAD(&sma->pending_alter); | |
225 | } | |
226 | ||
227 | /** | |
228 | * merge_queues - Merge single semop queues into global queue | |
229 | * @sma: semaphore array | |
230 | * | |
231 | * This function merges all per-semaphore queues into the global queue. | |
232 | * It is necessary to achieve FIFO ordering for the pending single-sop | |
233 | * operations when a multi-semop operation must sleep. | |
234 | * Only the alter operations must be moved, the const operations can stay. | |
235 | */ | |
236 | static void merge_queues(struct sem_array *sma) | |
237 | { | |
238 | int i; | |
239 | for (i = 0; i < sma->sem_nsems; i++) { | |
240 | struct sem *sem = sma->sem_base + i; | |
241 | ||
242 | list_splice_init(&sem->pending_alter, &sma->pending_alter); | |
243 | } | |
244 | } | |
245 | ||
e84ca333 DB |
246 | static void sem_rcu_free(struct rcu_head *head) |
247 | { | |
248 | struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu); | |
249 | struct sem_array *sma = ipc_rcu_to_struct(p); | |
250 | ||
251 | security_sem_free(sma); | |
252 | ipc_rcu_free(head); | |
253 | } | |
254 | ||
30e5bc30 MS |
255 | /* |
256 | * spin_unlock_wait() and !spin_is_locked() are not memory barriers, they | |
257 | * are only control barriers. | |
258 | * The code must pair with spin_unlock(&sem->lock) or | |
259 | * spin_unlock(&sem_perm.lock), thus just the control barrier is insufficient. | |
260 | * | |
261 | * smp_rmb() is sufficient, as writes cannot pass the control barrier. | |
262 | */ | |
263 | #define ipc_smp_acquire__after_spin_is_unlocked() smp_rmb() | |
264 | ||
184076a9 MS |
265 | /* |
266 | * Wait until all currently ongoing simple ops have completed. | |
267 | * Caller must own sem_perm.lock. | |
268 | * New simple ops cannot start, because simple ops first check | |
269 | * that sem_perm.lock is free. | |
901f6fed | 270 | * that a) sem_perm.lock is free and b) complex_count is 0. |
184076a9 MS |
271 | */ |
272 | static void sem_wait_array(struct sem_array *sma) | |
273 | { | |
274 | int i; | |
275 | struct sem *sem; | |
276 | ||
901f6fed MS |
277 | if (sma->complex_count) { |
278 | /* The thread that increased sma->complex_count waited on | |
279 | * all sem->lock locks. Thus we don't need to wait again. | |
280 | */ | |
281 | return; | |
282 | } | |
283 | ||
184076a9 MS |
284 | for (i = 0; i < sma->sem_nsems; i++) { |
285 | sem = sma->sem_base + i; | |
286 | spin_unlock_wait(&sem->lock); | |
287 | } | |
30e5bc30 | 288 | ipc_smp_acquire__after_spin_is_unlocked(); |
184076a9 MS |
289 | } |
290 | ||
6062a8dc RR |
291 | /* |
292 | * If the request contains only one semaphore operation, and there are | |
293 | * no complex transactions pending, lock only the semaphore involved. | |
294 | * Otherwise, lock the entire semaphore array, since we either have | |
295 | * multiple semaphores in our own semops, or we need to look at | |
296 | * semaphores from other pending complex operations. | |
6062a8dc RR |
297 | */ |
298 | static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, | |
299 | int nsops) | |
300 | { | |
184076a9 | 301 | struct sem *sem; |
6062a8dc | 302 | |
184076a9 MS |
303 | if (nsops != 1) { |
304 | /* Complex operation - acquire a full lock */ | |
305 | ipc_lock_object(&sma->sem_perm); | |
6062a8dc | 306 | |
184076a9 MS |
307 | /* And wait until all simple ops that are processed |
308 | * right now have dropped their locks. | |
6062a8dc | 309 | */ |
184076a9 MS |
310 | sem_wait_array(sma); |
311 | return -1; | |
312 | } | |
313 | ||
314 | /* | |
315 | * Only one semaphore affected - try to optimize locking. | |
316 | * The rules are: | |
317 | * - optimized locking is possible if no complex operation | |
318 | * is either enqueued or processed right now. | |
319 | * - The test for enqueued complex ops is simple: | |
320 | * sma->complex_count != 0 | |
321 | * - Testing for complex ops that are processed right now is | |
322 | * a bit more difficult. Complex ops acquire the full lock | |
323 | * and first wait that the running simple ops have completed. | |
324 | * (see above) | |
325 | * Thus: If we own a simple lock and the global lock is free | |
326 | * and complex_count is now 0, then it will stay 0 and | |
327 | * thus just locking sem->lock is sufficient. | |
328 | */ | |
329 | sem = sma->sem_base + sops->sem_num; | |
6062a8dc | 330 | |
184076a9 | 331 | if (sma->complex_count == 0) { |
6062a8dc | 332 | /* |
184076a9 MS |
333 | * It appears that no complex operation is around. |
334 | * Acquire the per-semaphore lock. | |
6062a8dc | 335 | */ |
184076a9 MS |
336 | spin_lock(&sem->lock); |
337 | ||
338 | /* Then check that the global lock is free */ | |
339 | if (!spin_is_locked(&sma->sem_perm.lock)) { | |
30e5bc30 MS |
340 | /* |
341 | * We need a memory barrier with acquire semantics, | |
342 | * otherwise we can race with another thread that does: | |
343 | * complex_count++; | |
344 | * spin_unlock(sem_perm.lock); | |
345 | */ | |
346 | ipc_smp_acquire__after_spin_is_unlocked(); | |
184076a9 MS |
347 | |
348 | /* Now repeat the test of complex_count: | |
349 | * It can't change anymore until we drop sem->lock. | |
350 | * Thus: if is now 0, then it will stay 0. | |
351 | */ | |
352 | if (sma->complex_count == 0) { | |
353 | /* fast path successful! */ | |
354 | return sops->sem_num; | |
355 | } | |
6062a8dc | 356 | } |
184076a9 MS |
357 | spin_unlock(&sem->lock); |
358 | } | |
359 | ||
360 | /* slow path: acquire the full lock */ | |
361 | ipc_lock_object(&sma->sem_perm); | |
6062a8dc | 362 | |
184076a9 MS |
363 | if (sma->complex_count == 0) { |
364 | /* False alarm: | |
365 | * There is no complex operation, thus we can switch | |
366 | * back to the fast path. | |
367 | */ | |
368 | spin_lock(&sem->lock); | |
369 | ipc_unlock_object(&sma->sem_perm); | |
370 | return sops->sem_num; | |
6062a8dc | 371 | } else { |
184076a9 MS |
372 | /* Not a false alarm, thus complete the sequence for a |
373 | * full lock. | |
6062a8dc | 374 | */ |
184076a9 MS |
375 | sem_wait_array(sma); |
376 | return -1; | |
6062a8dc | 377 | } |
6062a8dc RR |
378 | } |
379 | ||
380 | static inline void sem_unlock(struct sem_array *sma, int locknum) | |
381 | { | |
382 | if (locknum == -1) { | |
e5639c52 | 383 | unmerge_queues(sma); |
115d40db | 384 | ipc_unlock_object(&sma->sem_perm); |
6062a8dc RR |
385 | } else { |
386 | struct sem *sem = sma->sem_base + locknum; | |
387 | spin_unlock(&sem->lock); | |
388 | } | |
6062a8dc RR |
389 | } |
390 | ||
3e148c79 | 391 | /* |
33b74669 | 392 | * sem_lock_(check_) routines are called in the paths where the rwsem |
3e148c79 | 393 | * is not held. |
321310ce LT |
394 | * |
395 | * The caller holds the RCU read lock. | |
3e148c79 | 396 | */ |
6062a8dc RR |
397 | static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns, |
398 | int id, struct sembuf *sops, int nsops, int *locknum) | |
023a5355 | 399 | { |
c460b662 RR |
400 | struct kern_ipc_perm *ipcp; |
401 | struct sem_array *sma; | |
03f02c76 | 402 | |
c460b662 | 403 | ipcp = ipc_obtain_object(&sem_ids(ns), id); |
321310ce LT |
404 | if (IS_ERR(ipcp)) |
405 | return ERR_CAST(ipcp); | |
b1ed88b4 | 406 | |
6062a8dc RR |
407 | sma = container_of(ipcp, struct sem_array, sem_perm); |
408 | *locknum = sem_lock(sma, sops, nsops); | |
c460b662 RR |
409 | |
410 | /* ipc_rmid() may have already freed the ID while sem_lock | |
411 | * was spinning: verify that the structure is still valid | |
412 | */ | |
413 | if (!ipcp->deleted) | |
414 | return container_of(ipcp, struct sem_array, sem_perm); | |
415 | ||
6062a8dc | 416 | sem_unlock(sma, *locknum); |
321310ce | 417 | return ERR_PTR(-EINVAL); |
023a5355 ND |
418 | } |
419 | ||
16df3674 DB |
420 | static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id) |
421 | { | |
422 | struct kern_ipc_perm *ipcp = ipc_obtain_object(&sem_ids(ns), id); | |
423 | ||
424 | if (IS_ERR(ipcp)) | |
425 | return ERR_CAST(ipcp); | |
426 | ||
427 | return container_of(ipcp, struct sem_array, sem_perm); | |
428 | } | |
429 | ||
16df3674 DB |
430 | static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns, |
431 | int id) | |
432 | { | |
433 | struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id); | |
434 | ||
435 | if (IS_ERR(ipcp)) | |
436 | return ERR_CAST(ipcp); | |
b1ed88b4 | 437 | |
03f02c76 | 438 | return container_of(ipcp, struct sem_array, sem_perm); |
023a5355 ND |
439 | } |
440 | ||
6ff37972 PP |
441 | static inline void sem_lock_and_putref(struct sem_array *sma) |
442 | { | |
6062a8dc | 443 | sem_lock(sma, NULL, -1); |
e84ca333 | 444 | ipc_rcu_putref(sma, ipc_rcu_free); |
6ff37972 PP |
445 | } |
446 | ||
7ca7e564 ND |
447 | static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s) |
448 | { | |
449 | ipc_rmid(&sem_ids(ns), &s->sem_perm); | |
450 | } | |
451 | ||
1da177e4 LT |
452 | /* |
453 | * Lockless wakeup algorithm: | |
454 | * Without the check/retry algorithm a lockless wakeup is possible: | |
455 | * - queue.status is initialized to -EINTR before blocking. | |
456 | * - wakeup is performed by | |
ab63bc97 | 457 | * * unlinking the queue entry from the pending list |
1da177e4 LT |
458 | * * setting queue.status to IN_WAKEUP |
459 | * This is the notification for the blocked thread that a | |
460 | * result value is imminent. | |
461 | * * call wake_up_process | |
462 | * * set queue.status to the final value. | |
463 | * - the previously blocked thread checks queue.status: | |
464 | * * if it's IN_WAKEUP, then it must wait until the value changes | |
465 | * * if it's not -EINTR, then the operation was completed by | |
466 | * update_queue. semtimedop can return queue.status without | |
5f921ae9 | 467 | * performing any operation on the sem array. |
1da177e4 LT |
468 | * * otherwise it must acquire the spinlock and check what's up. |
469 | * | |
470 | * The two-stage algorithm is necessary to protect against the following | |
471 | * races: | |
472 | * - if queue.status is set after wake_up_process, then the woken up idle | |
473 | * thread could race forward and try (and fail) to acquire sma->lock | |
474 | * before update_queue had a chance to set queue.status | |
475 | * - if queue.status is written before wake_up_process and if the | |
476 | * blocked process is woken up by a signal between writing | |
477 | * queue.status and the wake_up_process, then the woken up | |
478 | * process could return from semtimedop and die by calling | |
479 | * sys_exit before wake_up_process is called. Then wake_up_process | |
480 | * will oops, because the task structure is already invalid. | |
481 | * (yes, this happened on s390 with sysv msg). | |
482 | * | |
483 | */ | |
484 | #define IN_WAKEUP 1 | |
485 | ||
f4566f04 ND |
486 | /** |
487 | * newary - Create a new semaphore set | |
488 | * @ns: namespace | |
489 | * @params: ptr to the structure that contains key, semflg and nsems | |
490 | * | |
33b74669 | 491 | * Called with sem_ids.rwsem held (as a writer) |
f4566f04 ND |
492 | */ |
493 | ||
7748dbfa | 494 | static int newary(struct ipc_namespace *ns, struct ipc_params *params) |
1da177e4 LT |
495 | { |
496 | int id; | |
497 | int retval; | |
498 | struct sem_array *sma; | |
499 | int size; | |
7748dbfa ND |
500 | key_t key = params->key; |
501 | int nsems = params->u.nsems; | |
502 | int semflg = params->flg; | |
b97e820f | 503 | int i; |
1da177e4 LT |
504 | |
505 | if (!nsems) | |
506 | return -EINVAL; | |
e3893534 | 507 | if (ns->used_sems + nsems > ns->sc_semmns) |
1da177e4 LT |
508 | return -ENOSPC; |
509 | ||
510 | size = sizeof (*sma) + nsems * sizeof (struct sem); | |
511 | sma = ipc_rcu_alloc(size); | |
512 | if (!sma) { | |
513 | return -ENOMEM; | |
514 | } | |
515 | memset (sma, 0, size); | |
516 | ||
517 | sma->sem_perm.mode = (semflg & S_IRWXUGO); | |
518 | sma->sem_perm.key = key; | |
519 | ||
520 | sma->sem_perm.security = NULL; | |
521 | retval = security_sem_alloc(sma); | |
522 | if (retval) { | |
e84ca333 | 523 | ipc_rcu_putref(sma, ipc_rcu_free); |
1da177e4 LT |
524 | return retval; |
525 | } | |
526 | ||
e3893534 | 527 | id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni); |
283bb7fa | 528 | if (id < 0) { |
e84ca333 | 529 | ipc_rcu_putref(sma, sem_rcu_free); |
283bb7fa | 530 | return id; |
1da177e4 | 531 | } |
e3893534 | 532 | ns->used_sems += nsems; |
1da177e4 LT |
533 | |
534 | sma->sem_base = (struct sem *) &sma[1]; | |
b97e820f | 535 | |
6062a8dc | 536 | for (i = 0; i < nsems; i++) { |
ab63bc97 MS |
537 | INIT_LIST_HEAD(&sma->sem_base[i].pending_alter); |
538 | INIT_LIST_HEAD(&sma->sem_base[i].pending_const); | |
6062a8dc RR |
539 | spin_lock_init(&sma->sem_base[i].lock); |
540 | } | |
b97e820f MS |
541 | |
542 | sma->complex_count = 0; | |
ab63bc97 MS |
543 | INIT_LIST_HEAD(&sma->pending_alter); |
544 | INIT_LIST_HEAD(&sma->pending_const); | |
4daa28f6 | 545 | INIT_LIST_HEAD(&sma->list_id); |
1da177e4 LT |
546 | sma->sem_nsems = nsems; |
547 | sma->sem_ctime = get_seconds(); | |
6062a8dc | 548 | sem_unlock(sma, -1); |
6d49dab8 | 549 | rcu_read_unlock(); |
1da177e4 | 550 | |
7ca7e564 | 551 | return sma->sem_perm.id; |
1da177e4 LT |
552 | } |
553 | ||
7748dbfa | 554 | |
f4566f04 | 555 | /* |
33b74669 | 556 | * Called with sem_ids.rwsem and ipcp locked. |
f4566f04 | 557 | */ |
03f02c76 | 558 | static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg) |
7748dbfa | 559 | { |
03f02c76 ND |
560 | struct sem_array *sma; |
561 | ||
562 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
563 | return security_sem_associate(sma, semflg); | |
7748dbfa ND |
564 | } |
565 | ||
f4566f04 | 566 | /* |
33b74669 | 567 | * Called with sem_ids.rwsem and ipcp locked. |
f4566f04 | 568 | */ |
03f02c76 ND |
569 | static inline int sem_more_checks(struct kern_ipc_perm *ipcp, |
570 | struct ipc_params *params) | |
7748dbfa | 571 | { |
03f02c76 ND |
572 | struct sem_array *sma; |
573 | ||
574 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
575 | if (params->u.nsems > sma->sem_nsems) | |
7748dbfa ND |
576 | return -EINVAL; |
577 | ||
578 | return 0; | |
579 | } | |
580 | ||
d5460c99 | 581 | SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg) |
1da177e4 | 582 | { |
e3893534 | 583 | struct ipc_namespace *ns; |
7748dbfa ND |
584 | struct ipc_ops sem_ops; |
585 | struct ipc_params sem_params; | |
e3893534 KK |
586 | |
587 | ns = current->nsproxy->ipc_ns; | |
1da177e4 | 588 | |
e3893534 | 589 | if (nsems < 0 || nsems > ns->sc_semmsl) |
1da177e4 | 590 | return -EINVAL; |
7ca7e564 | 591 | |
7748dbfa ND |
592 | sem_ops.getnew = newary; |
593 | sem_ops.associate = sem_security; | |
594 | sem_ops.more_checks = sem_more_checks; | |
595 | ||
596 | sem_params.key = key; | |
597 | sem_params.flg = semflg; | |
598 | sem_params.u.nsems = nsems; | |
1da177e4 | 599 | |
7748dbfa | 600 | return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params); |
1da177e4 LT |
601 | } |
602 | ||
b56e88e2 MS |
603 | /** perform_atomic_semop - Perform (if possible) a semaphore operation |
604 | * @sma: semaphore array | |
605 | * @sops: array with operations that should be checked | |
606 | * @nsems: number of sops | |
607 | * @un: undo array | |
608 | * @pid: pid that did the change | |
609 | * | |
610 | * Returns 0 if the operation was possible. | |
611 | * Returns 1 if the operation is impossible, the caller must sleep. | |
612 | * Negative values are error codes. | |
1da177e4 LT |
613 | */ |
614 | ||
b56e88e2 | 615 | static int perform_atomic_semop(struct sem_array *sma, struct sembuf *sops, |
1da177e4 LT |
616 | int nsops, struct sem_undo *un, int pid) |
617 | { | |
618 | int result, sem_op; | |
619 | struct sembuf *sop; | |
620 | struct sem * curr; | |
621 | ||
622 | for (sop = sops; sop < sops + nsops; sop++) { | |
623 | curr = sma->sem_base + sop->sem_num; | |
624 | sem_op = sop->sem_op; | |
625 | result = curr->semval; | |
626 | ||
627 | if (!sem_op && result) | |
628 | goto would_block; | |
629 | ||
630 | result += sem_op; | |
631 | if (result < 0) | |
632 | goto would_block; | |
633 | if (result > SEMVMX) | |
634 | goto out_of_range; | |
635 | if (sop->sem_flg & SEM_UNDO) { | |
636 | int undo = un->semadj[sop->sem_num] - sem_op; | |
637 | /* | |
638 | * Exceeding the undo range is an error. | |
639 | */ | |
640 | if (undo < (-SEMAEM - 1) || undo > SEMAEM) | |
641 | goto out_of_range; | |
642 | } | |
643 | curr->semval = result; | |
644 | } | |
645 | ||
646 | sop--; | |
647 | while (sop >= sops) { | |
648 | sma->sem_base[sop->sem_num].sempid = pid; | |
649 | if (sop->sem_flg & SEM_UNDO) | |
650 | un->semadj[sop->sem_num] -= sop->sem_op; | |
651 | sop--; | |
652 | } | |
653 | ||
1da177e4 LT |
654 | return 0; |
655 | ||
656 | out_of_range: | |
657 | result = -ERANGE; | |
658 | goto undo; | |
659 | ||
660 | would_block: | |
661 | if (sop->sem_flg & IPC_NOWAIT) | |
662 | result = -EAGAIN; | |
663 | else | |
664 | result = 1; | |
665 | ||
666 | undo: | |
667 | sop--; | |
668 | while (sop >= sops) { | |
669 | sma->sem_base[sop->sem_num].semval -= sop->sem_op; | |
670 | sop--; | |
671 | } | |
672 | ||
673 | return result; | |
674 | } | |
675 | ||
0a2b9d4c MS |
676 | /** wake_up_sem_queue_prepare(q, error): Prepare wake-up |
677 | * @q: queue entry that must be signaled | |
678 | * @error: Error value for the signal | |
679 | * | |
680 | * Prepare the wake-up of the queue entry q. | |
d4212093 | 681 | */ |
0a2b9d4c MS |
682 | static void wake_up_sem_queue_prepare(struct list_head *pt, |
683 | struct sem_queue *q, int error) | |
d4212093 | 684 | { |
0a2b9d4c MS |
685 | if (list_empty(pt)) { |
686 | /* | |
687 | * Hold preempt off so that we don't get preempted and have the | |
688 | * wakee busy-wait until we're scheduled back on. | |
689 | */ | |
690 | preempt_disable(); | |
691 | } | |
d4212093 | 692 | q->status = IN_WAKEUP; |
0a2b9d4c MS |
693 | q->pid = error; |
694 | ||
9f1bc2c9 | 695 | list_add_tail(&q->list, pt); |
0a2b9d4c MS |
696 | } |
697 | ||
698 | /** | |
699 | * wake_up_sem_queue_do(pt) - do the actual wake-up | |
700 | * @pt: list of tasks to be woken up | |
701 | * | |
702 | * Do the actual wake-up. | |
703 | * The function is called without any locks held, thus the semaphore array | |
704 | * could be destroyed already and the tasks can disappear as soon as the | |
705 | * status is set to the actual return code. | |
706 | */ | |
707 | static void wake_up_sem_queue_do(struct list_head *pt) | |
708 | { | |
709 | struct sem_queue *q, *t; | |
710 | int did_something; | |
711 | ||
712 | did_something = !list_empty(pt); | |
9f1bc2c9 | 713 | list_for_each_entry_safe(q, t, pt, list) { |
0a2b9d4c MS |
714 | wake_up_process(q->sleeper); |
715 | /* q can disappear immediately after writing q->status. */ | |
716 | smp_wmb(); | |
717 | q->status = q->pid; | |
718 | } | |
719 | if (did_something) | |
720 | preempt_enable(); | |
d4212093 NP |
721 | } |
722 | ||
b97e820f MS |
723 | static void unlink_queue(struct sem_array *sma, struct sem_queue *q) |
724 | { | |
725 | list_del(&q->list); | |
9f1bc2c9 | 726 | if (q->nsops > 1) |
b97e820f MS |
727 | sma->complex_count--; |
728 | } | |
729 | ||
fd5db422 MS |
730 | /** check_restart(sma, q) |
731 | * @sma: semaphore array | |
732 | * @q: the operation that just completed | |
733 | * | |
734 | * update_queue is O(N^2) when it restarts scanning the whole queue of | |
735 | * waiting operations. Therefore this function checks if the restart is | |
736 | * really necessary. It is called after a previously waiting operation | |
ab63bc97 MS |
737 | * modified the array. |
738 | * Note that wait-for-zero operations are handled without restart. | |
fd5db422 MS |
739 | */ |
740 | static int check_restart(struct sem_array *sma, struct sem_queue *q) | |
741 | { | |
ab63bc97 MS |
742 | /* pending complex alter operations are too difficult to analyse */ |
743 | if (!list_empty(&sma->pending_alter)) | |
fd5db422 MS |
744 | return 1; |
745 | ||
746 | /* we were a sleeping complex operation. Too difficult */ | |
747 | if (q->nsops > 1) | |
748 | return 1; | |
749 | ||
ab63bc97 MS |
750 | /* It is impossible that someone waits for the new value: |
751 | * - complex operations always restart. | |
752 | * - wait-for-zero are handled seperately. | |
753 | * - q is a previously sleeping simple operation that | |
754 | * altered the array. It must be a decrement, because | |
755 | * simple increments never sleep. | |
756 | * - If there are older (higher priority) decrements | |
757 | * in the queue, then they have observed the original | |
758 | * semval value and couldn't proceed. The operation | |
759 | * decremented to value - thus they won't proceed either. | |
760 | */ | |
761 | return 0; | |
762 | } | |
fd5db422 | 763 | |
ab63bc97 MS |
764 | /** |
765 | * wake_const_ops(sma, semnum, pt) - Wake up non-alter tasks | |
766 | * @sma: semaphore array. | |
767 | * @semnum: semaphore that was modified. | |
768 | * @pt: list head for the tasks that must be woken up. | |
769 | * | |
770 | * wake_const_ops must be called after a semaphore in a semaphore array | |
771 | * was set to 0. If complex const operations are pending, wake_const_ops must | |
772 | * be called with semnum = -1, as well as with the number of each modified | |
773 | * semaphore. | |
774 | * The tasks that must be woken up are added to @pt. The return code | |
775 | * is stored in q->pid. | |
776 | * The function returns 1 if at least one operation was completed successfully. | |
777 | */ | |
778 | static int wake_const_ops(struct sem_array *sma, int semnum, | |
779 | struct list_head *pt) | |
780 | { | |
781 | struct sem_queue *q; | |
782 | struct list_head *walk; | |
783 | struct list_head *pending_list; | |
784 | int semop_completed = 0; | |
785 | ||
786 | if (semnum == -1) | |
787 | pending_list = &sma->pending_const; | |
788 | else | |
789 | pending_list = &sma->sem_base[semnum].pending_const; | |
fd5db422 | 790 | |
ab63bc97 MS |
791 | walk = pending_list->next; |
792 | while (walk != pending_list) { | |
793 | int error; | |
794 | ||
795 | q = container_of(walk, struct sem_queue, list); | |
796 | walk = walk->next; | |
797 | ||
b56e88e2 MS |
798 | error = perform_atomic_semop(sma, q->sops, q->nsops, |
799 | q->undo, q->pid); | |
ab63bc97 MS |
800 | |
801 | if (error <= 0) { | |
802 | /* operation completed, remove from queue & wakeup */ | |
803 | ||
804 | unlink_queue(sma, q); | |
805 | ||
806 | wake_up_sem_queue_prepare(pt, q, error); | |
807 | if (error == 0) | |
808 | semop_completed = 1; | |
809 | } | |
810 | } | |
811 | return semop_completed; | |
812 | } | |
813 | ||
814 | /** | |
815 | * do_smart_wakeup_zero(sma, sops, nsops, pt) - wakeup all wait for zero tasks | |
816 | * @sma: semaphore array | |
817 | * @sops: operations that were performed | |
818 | * @nsops: number of operations | |
819 | * @pt: list head of the tasks that must be woken up. | |
820 | * | |
821 | * do_smart_wakeup_zero() checks all required queue for wait-for-zero | |
822 | * operations, based on the actual changes that were performed on the | |
823 | * semaphore array. | |
824 | * The function returns 1 if at least one operation was completed successfully. | |
825 | */ | |
826 | static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops, | |
827 | int nsops, struct list_head *pt) | |
828 | { | |
829 | int i; | |
830 | int semop_completed = 0; | |
831 | int got_zero = 0; | |
832 | ||
833 | /* first: the per-semaphore queues, if known */ | |
834 | if (sops) { | |
835 | for (i = 0; i < nsops; i++) { | |
836 | int num = sops[i].sem_num; | |
837 | ||
838 | if (sma->sem_base[num].semval == 0) { | |
839 | got_zero = 1; | |
840 | semop_completed |= wake_const_ops(sma, num, pt); | |
841 | } | |
842 | } | |
843 | } else { | |
844 | /* | |
845 | * No sops means modified semaphores not known. | |
846 | * Assume all were changed. | |
fd5db422 | 847 | */ |
ab63bc97 MS |
848 | for (i = 0; i < sma->sem_nsems; i++) { |
849 | if (sma->sem_base[i].semval == 0) { | |
850 | got_zero = 1; | |
851 | semop_completed |= wake_const_ops(sma, i, pt); | |
852 | } | |
853 | } | |
fd5db422 MS |
854 | } |
855 | /* | |
ab63bc97 MS |
856 | * If one of the modified semaphores got 0, |
857 | * then check the global queue, too. | |
fd5db422 | 858 | */ |
ab63bc97 MS |
859 | if (got_zero) |
860 | semop_completed |= wake_const_ops(sma, -1, pt); | |
fd5db422 | 861 | |
ab63bc97 | 862 | return semop_completed; |
fd5db422 MS |
863 | } |
864 | ||
636c6be8 MS |
865 | |
866 | /** | |
867 | * update_queue(sma, semnum): Look for tasks that can be completed. | |
868 | * @sma: semaphore array. | |
869 | * @semnum: semaphore that was modified. | |
0a2b9d4c | 870 | * @pt: list head for the tasks that must be woken up. |
636c6be8 MS |
871 | * |
872 | * update_queue must be called after a semaphore in a semaphore array | |
9f1bc2c9 RR |
873 | * was modified. If multiple semaphores were modified, update_queue must |
874 | * be called with semnum = -1, as well as with the number of each modified | |
875 | * semaphore. | |
0a2b9d4c MS |
876 | * The tasks that must be woken up are added to @pt. The return code |
877 | * is stored in q->pid. | |
ab63bc97 MS |
878 | * The function internally checks if const operations can now succeed. |
879 | * | |
0a2b9d4c | 880 | * The function return 1 if at least one semop was completed successfully. |
1da177e4 | 881 | */ |
0a2b9d4c | 882 | static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt) |
1da177e4 | 883 | { |
636c6be8 MS |
884 | struct sem_queue *q; |
885 | struct list_head *walk; | |
886 | struct list_head *pending_list; | |
0a2b9d4c | 887 | int semop_completed = 0; |
636c6be8 | 888 | |
9f1bc2c9 | 889 | if (semnum == -1) |
ab63bc97 | 890 | pending_list = &sma->pending_alter; |
9f1bc2c9 | 891 | else |
ab63bc97 | 892 | pending_list = &sma->sem_base[semnum].pending_alter; |
9cad200c NP |
893 | |
894 | again: | |
636c6be8 MS |
895 | walk = pending_list->next; |
896 | while (walk != pending_list) { | |
fd5db422 | 897 | int error, restart; |
636c6be8 | 898 | |
9f1bc2c9 | 899 | q = container_of(walk, struct sem_queue, list); |
636c6be8 | 900 | walk = walk->next; |
1da177e4 | 901 | |
d987f8b2 MS |
902 | /* If we are scanning the single sop, per-semaphore list of |
903 | * one semaphore and that semaphore is 0, then it is not | |
ab63bc97 | 904 | * necessary to scan further: simple increments |
d987f8b2 MS |
905 | * that affect only one entry succeed immediately and cannot |
906 | * be in the per semaphore pending queue, and decrements | |
907 | * cannot be successful if the value is already 0. | |
908 | */ | |
ab63bc97 | 909 | if (semnum != -1 && sma->sem_base[semnum].semval == 0) |
d987f8b2 MS |
910 | break; |
911 | ||
b56e88e2 | 912 | error = perform_atomic_semop(sma, q->sops, q->nsops, |
1da177e4 LT |
913 | q->undo, q->pid); |
914 | ||
915 | /* Does q->sleeper still need to sleep? */ | |
9cad200c NP |
916 | if (error > 0) |
917 | continue; | |
918 | ||
b97e820f | 919 | unlink_queue(sma, q); |
9cad200c | 920 | |
0a2b9d4c | 921 | if (error) { |
fd5db422 | 922 | restart = 0; |
0a2b9d4c MS |
923 | } else { |
924 | semop_completed = 1; | |
ab63bc97 | 925 | do_smart_wakeup_zero(sma, q->sops, q->nsops, pt); |
fd5db422 | 926 | restart = check_restart(sma, q); |
0a2b9d4c | 927 | } |
fd5db422 | 928 | |
0a2b9d4c | 929 | wake_up_sem_queue_prepare(pt, q, error); |
fd5db422 | 930 | if (restart) |
9cad200c | 931 | goto again; |
1da177e4 | 932 | } |
0a2b9d4c | 933 | return semop_completed; |
1da177e4 LT |
934 | } |
935 | ||
e556ea01 MS |
936 | /** |
937 | * set_semotime(sma, sops) - set sem_otime | |
938 | * @sma: semaphore array | |
939 | * @sops: operations that modified the array, may be NULL | |
940 | * | |
941 | * sem_otime is replicated to avoid cache line trashing. | |
942 | * This function sets one instance to the current time. | |
943 | */ | |
944 | static void set_semotime(struct sem_array *sma, struct sembuf *sops) | |
945 | { | |
946 | if (sops == NULL) { | |
947 | sma->sem_base[0].sem_otime = get_seconds(); | |
948 | } else { | |
949 | sma->sem_base[sops[0].sem_num].sem_otime = | |
950 | get_seconds(); | |
951 | } | |
952 | } | |
953 | ||
0a2b9d4c MS |
954 | /** |
955 | * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue | |
fd5db422 MS |
956 | * @sma: semaphore array |
957 | * @sops: operations that were performed | |
958 | * @nsops: number of operations | |
0a2b9d4c MS |
959 | * @otime: force setting otime |
960 | * @pt: list head of the tasks that must be woken up. | |
fd5db422 | 961 | * |
ab63bc97 MS |
962 | * do_smart_update() does the required calls to update_queue and wakeup_zero, |
963 | * based on the actual changes that were performed on the semaphore array. | |
0a2b9d4c MS |
964 | * Note that the function does not do the actual wake-up: the caller is |
965 | * responsible for calling wake_up_sem_queue_do(@pt). | |
966 | * It is safe to perform this call after dropping all locks. | |
fd5db422 | 967 | */ |
0a2b9d4c MS |
968 | static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops, |
969 | int otime, struct list_head *pt) | |
fd5db422 MS |
970 | { |
971 | int i; | |
972 | ||
ab63bc97 MS |
973 | otime |= do_smart_wakeup_zero(sma, sops, nsops, pt); |
974 | ||
e5639c52 MS |
975 | if (!list_empty(&sma->pending_alter)) { |
976 | /* semaphore array uses the global queue - just process it. */ | |
977 | otime |= update_queue(sma, -1, pt); | |
978 | } else { | |
979 | if (!sops) { | |
980 | /* | |
981 | * No sops, thus the modified semaphores are not | |
982 | * known. Check all. | |
983 | */ | |
984 | for (i = 0; i < sma->sem_nsems; i++) | |
985 | otime |= update_queue(sma, i, pt); | |
986 | } else { | |
987 | /* | |
988 | * Check the semaphores that were increased: | |
989 | * - No complex ops, thus all sleeping ops are | |
990 | * decrease. | |
991 | * - if we decreased the value, then any sleeping | |
992 | * semaphore ops wont be able to run: If the | |
993 | * previous value was too small, then the new | |
994 | * value will be too small, too. | |
995 | */ | |
996 | for (i = 0; i < nsops; i++) { | |
997 | if (sops[i].sem_op > 0) { | |
998 | otime |= update_queue(sma, | |
999 | sops[i].sem_num, pt); | |
1000 | } | |
ab465df9 | 1001 | } |
9f1bc2c9 | 1002 | } |
fd5db422 | 1003 | } |
e556ea01 MS |
1004 | if (otime) |
1005 | set_semotime(sma, sops); | |
fd5db422 MS |
1006 | } |
1007 | ||
1da177e4 LT |
1008 | /* The following counts are associated to each semaphore: |
1009 | * semncnt number of tasks waiting on semval being nonzero | |
1010 | * semzcnt number of tasks waiting on semval being zero | |
1011 | * This model assumes that a task waits on exactly one semaphore. | |
1012 | * Since semaphore operations are to be performed atomically, tasks actually | |
1013 | * wait on a whole sequence of semaphores simultaneously. | |
1014 | * The counts we return here are a rough approximation, but still | |
1015 | * warrant that semncnt+semzcnt>0 if the task is on the pending queue. | |
1016 | */ | |
1017 | static int count_semncnt (struct sem_array * sma, ushort semnum) | |
1018 | { | |
1019 | int semncnt; | |
1020 | struct sem_queue * q; | |
1021 | ||
1022 | semncnt = 0; | |
ab63bc97 | 1023 | list_for_each_entry(q, &sma->sem_base[semnum].pending_alter, list) { |
de2657f9 RR |
1024 | struct sembuf * sops = q->sops; |
1025 | BUG_ON(sops->sem_num != semnum); | |
1026 | if ((sops->sem_op < 0) && !(sops->sem_flg & IPC_NOWAIT)) | |
1027 | semncnt++; | |
1028 | } | |
1029 | ||
ab63bc97 | 1030 | list_for_each_entry(q, &sma->pending_alter, list) { |
1da177e4 LT |
1031 | struct sembuf * sops = q->sops; |
1032 | int nsops = q->nsops; | |
1033 | int i; | |
1034 | for (i = 0; i < nsops; i++) | |
1035 | if (sops[i].sem_num == semnum | |
1036 | && (sops[i].sem_op < 0) | |
1037 | && !(sops[i].sem_flg & IPC_NOWAIT)) | |
1038 | semncnt++; | |
1039 | } | |
1040 | return semncnt; | |
1041 | } | |
a1193f8e | 1042 | |
1da177e4 LT |
1043 | static int count_semzcnt (struct sem_array * sma, ushort semnum) |
1044 | { | |
1045 | int semzcnt; | |
1046 | struct sem_queue * q; | |
1047 | ||
1048 | semzcnt = 0; | |
ab63bc97 | 1049 | list_for_each_entry(q, &sma->sem_base[semnum].pending_const, list) { |
ebc2e5e6 RR |
1050 | struct sembuf * sops = q->sops; |
1051 | BUG_ON(sops->sem_num != semnum); | |
1052 | if ((sops->sem_op == 0) && !(sops->sem_flg & IPC_NOWAIT)) | |
1053 | semzcnt++; | |
1054 | } | |
1055 | ||
ab63bc97 | 1056 | list_for_each_entry(q, &sma->pending_const, list) { |
1da177e4 LT |
1057 | struct sembuf * sops = q->sops; |
1058 | int nsops = q->nsops; | |
1059 | int i; | |
1060 | for (i = 0; i < nsops; i++) | |
1061 | if (sops[i].sem_num == semnum | |
1062 | && (sops[i].sem_op == 0) | |
1063 | && !(sops[i].sem_flg & IPC_NOWAIT)) | |
1064 | semzcnt++; | |
1065 | } | |
1066 | return semzcnt; | |
1067 | } | |
1068 | ||
33b74669 DB |
1069 | /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked |
1070 | * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem | |
3e148c79 | 1071 | * remains locked on exit. |
1da177e4 | 1072 | */ |
01b8b07a | 1073 | static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp) |
1da177e4 | 1074 | { |
380af1b3 MS |
1075 | struct sem_undo *un, *tu; |
1076 | struct sem_queue *q, *tq; | |
01b8b07a | 1077 | struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm); |
0a2b9d4c | 1078 | struct list_head tasks; |
9f1bc2c9 | 1079 | int i; |
1da177e4 | 1080 | |
380af1b3 | 1081 | /* Free the existing undo structures for this semaphore set. */ |
115d40db | 1082 | ipc_assert_locked_object(&sma->sem_perm); |
380af1b3 MS |
1083 | list_for_each_entry_safe(un, tu, &sma->list_id, list_id) { |
1084 | list_del(&un->list_id); | |
1085 | spin_lock(&un->ulp->lock); | |
1da177e4 | 1086 | un->semid = -1; |
380af1b3 MS |
1087 | list_del_rcu(&un->list_proc); |
1088 | spin_unlock(&un->ulp->lock); | |
693a8b6e | 1089 | kfree_rcu(un, rcu); |
380af1b3 | 1090 | } |
1da177e4 LT |
1091 | |
1092 | /* Wake up all pending processes and let them fail with EIDRM. */ | |
0a2b9d4c | 1093 | INIT_LIST_HEAD(&tasks); |
ab63bc97 MS |
1094 | list_for_each_entry_safe(q, tq, &sma->pending_const, list) { |
1095 | unlink_queue(sma, q); | |
1096 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); | |
1097 | } | |
1098 | ||
1099 | list_for_each_entry_safe(q, tq, &sma->pending_alter, list) { | |
b97e820f | 1100 | unlink_queue(sma, q); |
0a2b9d4c | 1101 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); |
1da177e4 | 1102 | } |
9f1bc2c9 RR |
1103 | for (i = 0; i < sma->sem_nsems; i++) { |
1104 | struct sem *sem = sma->sem_base + i; | |
ab63bc97 MS |
1105 | list_for_each_entry_safe(q, tq, &sem->pending_const, list) { |
1106 | unlink_queue(sma, q); | |
1107 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); | |
1108 | } | |
1109 | list_for_each_entry_safe(q, tq, &sem->pending_alter, list) { | |
9f1bc2c9 RR |
1110 | unlink_queue(sma, q); |
1111 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); | |
1112 | } | |
1113 | } | |
1da177e4 | 1114 | |
7ca7e564 ND |
1115 | /* Remove the semaphore set from the IDR */ |
1116 | sem_rmid(ns, sma); | |
6062a8dc | 1117 | sem_unlock(sma, -1); |
6d49dab8 | 1118 | rcu_read_unlock(); |
1da177e4 | 1119 | |
0a2b9d4c | 1120 | wake_up_sem_queue_do(&tasks); |
e3893534 | 1121 | ns->used_sems -= sma->sem_nsems; |
e84ca333 | 1122 | ipc_rcu_putref(sma, sem_rcu_free); |
1da177e4 LT |
1123 | } |
1124 | ||
1125 | static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version) | |
1126 | { | |
1127 | switch(version) { | |
1128 | case IPC_64: | |
1129 | return copy_to_user(buf, in, sizeof(*in)); | |
1130 | case IPC_OLD: | |
1131 | { | |
1132 | struct semid_ds out; | |
1133 | ||
982f7c2b DR |
1134 | memset(&out, 0, sizeof(out)); |
1135 | ||
1da177e4 LT |
1136 | ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm); |
1137 | ||
1138 | out.sem_otime = in->sem_otime; | |
1139 | out.sem_ctime = in->sem_ctime; | |
1140 | out.sem_nsems = in->sem_nsems; | |
1141 | ||
1142 | return copy_to_user(buf, &out, sizeof(out)); | |
1143 | } | |
1144 | default: | |
1145 | return -EINVAL; | |
1146 | } | |
1147 | } | |
1148 | ||
bf6830ad MS |
1149 | static time_t get_semotime(struct sem_array *sma) |
1150 | { | |
1151 | int i; | |
1152 | time_t res; | |
1153 | ||
1154 | res = sma->sem_base[0].sem_otime; | |
1155 | for (i = 1; i < sma->sem_nsems; i++) { | |
1156 | time_t to = sma->sem_base[i].sem_otime; | |
1157 | ||
1158 | if (to > res) | |
1159 | res = to; | |
1160 | } | |
1161 | return res; | |
1162 | } | |
1163 | ||
4b9fcb0e | 1164 | static int semctl_nolock(struct ipc_namespace *ns, int semid, |
e1fd1f49 | 1165 | int cmd, int version, void __user *p) |
1da177e4 | 1166 | { |
e5cc9c7b | 1167 | int err; |
1da177e4 LT |
1168 | struct sem_array *sma; |
1169 | ||
1170 | switch(cmd) { | |
1171 | case IPC_INFO: | |
1172 | case SEM_INFO: | |
1173 | { | |
1174 | struct seminfo seminfo; | |
1175 | int max_id; | |
1176 | ||
1177 | err = security_sem_semctl(NULL, cmd); | |
1178 | if (err) | |
1179 | return err; | |
1180 | ||
1181 | memset(&seminfo,0,sizeof(seminfo)); | |
e3893534 KK |
1182 | seminfo.semmni = ns->sc_semmni; |
1183 | seminfo.semmns = ns->sc_semmns; | |
1184 | seminfo.semmsl = ns->sc_semmsl; | |
1185 | seminfo.semopm = ns->sc_semopm; | |
1da177e4 LT |
1186 | seminfo.semvmx = SEMVMX; |
1187 | seminfo.semmnu = SEMMNU; | |
1188 | seminfo.semmap = SEMMAP; | |
1189 | seminfo.semume = SEMUME; | |
33b74669 | 1190 | down_read(&sem_ids(ns).rwsem); |
1da177e4 | 1191 | if (cmd == SEM_INFO) { |
e3893534 KK |
1192 | seminfo.semusz = sem_ids(ns).in_use; |
1193 | seminfo.semaem = ns->used_sems; | |
1da177e4 LT |
1194 | } else { |
1195 | seminfo.semusz = SEMUSZ; | |
1196 | seminfo.semaem = SEMAEM; | |
1197 | } | |
7ca7e564 | 1198 | max_id = ipc_get_maxid(&sem_ids(ns)); |
33b74669 | 1199 | up_read(&sem_ids(ns).rwsem); |
e1fd1f49 | 1200 | if (copy_to_user(p, &seminfo, sizeof(struct seminfo))) |
1da177e4 LT |
1201 | return -EFAULT; |
1202 | return (max_id < 0) ? 0: max_id; | |
1203 | } | |
4b9fcb0e | 1204 | case IPC_STAT: |
1da177e4 LT |
1205 | case SEM_STAT: |
1206 | { | |
1207 | struct semid64_ds tbuf; | |
16df3674 DB |
1208 | int id = 0; |
1209 | ||
1210 | memset(&tbuf, 0, sizeof(tbuf)); | |
1da177e4 | 1211 | |
941b0304 | 1212 | rcu_read_lock(); |
4b9fcb0e | 1213 | if (cmd == SEM_STAT) { |
16df3674 DB |
1214 | sma = sem_obtain_object(ns, semid); |
1215 | if (IS_ERR(sma)) { | |
1216 | err = PTR_ERR(sma); | |
1217 | goto out_unlock; | |
1218 | } | |
4b9fcb0e PP |
1219 | id = sma->sem_perm.id; |
1220 | } else { | |
16df3674 DB |
1221 | sma = sem_obtain_object_check(ns, semid); |
1222 | if (IS_ERR(sma)) { | |
1223 | err = PTR_ERR(sma); | |
1224 | goto out_unlock; | |
1225 | } | |
4b9fcb0e | 1226 | } |
1da177e4 LT |
1227 | |
1228 | err = -EACCES; | |
b0e77598 | 1229 | if (ipcperms(ns, &sma->sem_perm, S_IRUGO)) |
1da177e4 LT |
1230 | goto out_unlock; |
1231 | ||
1232 | err = security_sem_semctl(sma, cmd); | |
1233 | if (err) | |
1234 | goto out_unlock; | |
1235 | ||
1da177e4 | 1236 | kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm); |
bf6830ad MS |
1237 | tbuf.sem_otime = get_semotime(sma); |
1238 | tbuf.sem_ctime = sma->sem_ctime; | |
1239 | tbuf.sem_nsems = sma->sem_nsems; | |
16df3674 | 1240 | rcu_read_unlock(); |
e1fd1f49 | 1241 | if (copy_semid_to_user(p, &tbuf, version)) |
1da177e4 LT |
1242 | return -EFAULT; |
1243 | return id; | |
1244 | } | |
1245 | default: | |
1246 | return -EINVAL; | |
1247 | } | |
1da177e4 | 1248 | out_unlock: |
16df3674 | 1249 | rcu_read_unlock(); |
1da177e4 LT |
1250 | return err; |
1251 | } | |
1252 | ||
e1fd1f49 AV |
1253 | static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum, |
1254 | unsigned long arg) | |
1255 | { | |
1256 | struct sem_undo *un; | |
1257 | struct sem_array *sma; | |
1258 | struct sem* curr; | |
1259 | int err; | |
e1fd1f49 AV |
1260 | struct list_head tasks; |
1261 | int val; | |
1262 | #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN) | |
1263 | /* big-endian 64bit */ | |
1264 | val = arg >> 32; | |
1265 | #else | |
1266 | /* 32bit or little-endian 64bit */ | |
1267 | val = arg; | |
1268 | #endif | |
1269 | ||
6062a8dc RR |
1270 | if (val > SEMVMX || val < 0) |
1271 | return -ERANGE; | |
e1fd1f49 AV |
1272 | |
1273 | INIT_LIST_HEAD(&tasks); | |
e1fd1f49 | 1274 | |
6062a8dc RR |
1275 | rcu_read_lock(); |
1276 | sma = sem_obtain_object_check(ns, semid); | |
1277 | if (IS_ERR(sma)) { | |
1278 | rcu_read_unlock(); | |
1279 | return PTR_ERR(sma); | |
1280 | } | |
1281 | ||
1282 | if (semnum < 0 || semnum >= sma->sem_nsems) { | |
1283 | rcu_read_unlock(); | |
1284 | return -EINVAL; | |
1285 | } | |
1286 | ||
1287 | ||
1288 | if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) { | |
1289 | rcu_read_unlock(); | |
1290 | return -EACCES; | |
1291 | } | |
e1fd1f49 AV |
1292 | |
1293 | err = security_sem_semctl(sma, SETVAL); | |
6062a8dc RR |
1294 | if (err) { |
1295 | rcu_read_unlock(); | |
1296 | return -EACCES; | |
1297 | } | |
e1fd1f49 | 1298 | |
6062a8dc | 1299 | sem_lock(sma, NULL, -1); |
e1fd1f49 | 1300 | |
873be93b MS |
1301 | if (sma->sem_perm.deleted) { |
1302 | sem_unlock(sma, -1); | |
1303 | rcu_read_unlock(); | |
1304 | return -EIDRM; | |
1305 | } | |
1306 | ||
e1fd1f49 AV |
1307 | curr = &sma->sem_base[semnum]; |
1308 | ||
115d40db | 1309 | ipc_assert_locked_object(&sma->sem_perm); |
e1fd1f49 AV |
1310 | list_for_each_entry(un, &sma->list_id, list_id) |
1311 | un->semadj[semnum] = 0; | |
1312 | ||
1313 | curr->semval = val; | |
1314 | curr->sempid = task_tgid_vnr(current); | |
1315 | sma->sem_ctime = get_seconds(); | |
1316 | /* maybe some queued-up processes were waiting for this */ | |
1317 | do_smart_update(sma, NULL, 0, 0, &tasks); | |
6062a8dc | 1318 | sem_unlock(sma, -1); |
6d49dab8 | 1319 | rcu_read_unlock(); |
e1fd1f49 | 1320 | wake_up_sem_queue_do(&tasks); |
6062a8dc | 1321 | return 0; |
e1fd1f49 AV |
1322 | } |
1323 | ||
e3893534 | 1324 | static int semctl_main(struct ipc_namespace *ns, int semid, int semnum, |
e1fd1f49 | 1325 | int cmd, void __user *p) |
1da177e4 LT |
1326 | { |
1327 | struct sem_array *sma; | |
1328 | struct sem* curr; | |
16df3674 | 1329 | int err, nsems; |
1da177e4 LT |
1330 | ushort fast_sem_io[SEMMSL_FAST]; |
1331 | ushort* sem_io = fast_sem_io; | |
0a2b9d4c | 1332 | struct list_head tasks; |
1da177e4 | 1333 | |
16df3674 DB |
1334 | INIT_LIST_HEAD(&tasks); |
1335 | ||
1336 | rcu_read_lock(); | |
1337 | sma = sem_obtain_object_check(ns, semid); | |
1338 | if (IS_ERR(sma)) { | |
1339 | rcu_read_unlock(); | |
023a5355 | 1340 | return PTR_ERR(sma); |
16df3674 | 1341 | } |
1da177e4 LT |
1342 | |
1343 | nsems = sma->sem_nsems; | |
1344 | ||
1da177e4 | 1345 | err = -EACCES; |
c728b9c8 LT |
1346 | if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO)) |
1347 | goto out_rcu_wakeup; | |
1da177e4 LT |
1348 | |
1349 | err = security_sem_semctl(sma, cmd); | |
c728b9c8 LT |
1350 | if (err) |
1351 | goto out_rcu_wakeup; | |
1da177e4 LT |
1352 | |
1353 | err = -EACCES; | |
1354 | switch (cmd) { | |
1355 | case GETALL: | |
1356 | { | |
e1fd1f49 | 1357 | ushort __user *array = p; |
1da177e4 LT |
1358 | int i; |
1359 | ||
ce857229 | 1360 | sem_lock(sma, NULL, -1); |
873be93b MS |
1361 | if (sma->sem_perm.deleted) { |
1362 | err = -EIDRM; | |
1363 | goto out_unlock; | |
1364 | } | |
1da177e4 | 1365 | if(nsems > SEMMSL_FAST) { |
ce857229 | 1366 | if (!ipc_rcu_getref(sma)) { |
ce857229 | 1367 | err = -EIDRM; |
873be93b | 1368 | goto out_unlock; |
ce857229 AV |
1369 | } |
1370 | sem_unlock(sma, -1); | |
6d49dab8 | 1371 | rcu_read_unlock(); |
1da177e4 LT |
1372 | sem_io = ipc_alloc(sizeof(ushort)*nsems); |
1373 | if(sem_io == NULL) { | |
e84ca333 | 1374 | ipc_rcu_putref(sma, ipc_rcu_free); |
1da177e4 LT |
1375 | return -ENOMEM; |
1376 | } | |
1377 | ||
4091fd94 | 1378 | rcu_read_lock(); |
6ff37972 | 1379 | sem_lock_and_putref(sma); |
1da177e4 | 1380 | if (sma->sem_perm.deleted) { |
1da177e4 | 1381 | err = -EIDRM; |
873be93b | 1382 | goto out_unlock; |
1da177e4 | 1383 | } |
ce857229 | 1384 | } |
1da177e4 LT |
1385 | for (i = 0; i < sma->sem_nsems; i++) |
1386 | sem_io[i] = sma->sem_base[i].semval; | |
6062a8dc | 1387 | sem_unlock(sma, -1); |
6d49dab8 | 1388 | rcu_read_unlock(); |
1da177e4 LT |
1389 | err = 0; |
1390 | if(copy_to_user(array, sem_io, nsems*sizeof(ushort))) | |
1391 | err = -EFAULT; | |
1392 | goto out_free; | |
1393 | } | |
1394 | case SETALL: | |
1395 | { | |
1396 | int i; | |
1397 | struct sem_undo *un; | |
1398 | ||
6062a8dc | 1399 | if (!ipc_rcu_getref(sma)) { |
873be93b MS |
1400 | err = -EIDRM; |
1401 | goto out_rcu_wakeup; | |
6062a8dc | 1402 | } |
16df3674 | 1403 | rcu_read_unlock(); |
1da177e4 LT |
1404 | |
1405 | if(nsems > SEMMSL_FAST) { | |
1406 | sem_io = ipc_alloc(sizeof(ushort)*nsems); | |
1407 | if(sem_io == NULL) { | |
e84ca333 | 1408 | ipc_rcu_putref(sma, ipc_rcu_free); |
1da177e4 LT |
1409 | return -ENOMEM; |
1410 | } | |
1411 | } | |
1412 | ||
e1fd1f49 | 1413 | if (copy_from_user (sem_io, p, nsems*sizeof(ushort))) { |
e84ca333 | 1414 | ipc_rcu_putref(sma, ipc_rcu_free); |
1da177e4 LT |
1415 | err = -EFAULT; |
1416 | goto out_free; | |
1417 | } | |
1418 | ||
1419 | for (i = 0; i < nsems; i++) { | |
1420 | if (sem_io[i] > SEMVMX) { | |
e84ca333 | 1421 | ipc_rcu_putref(sma, ipc_rcu_free); |
1da177e4 LT |
1422 | err = -ERANGE; |
1423 | goto out_free; | |
1424 | } | |
1425 | } | |
4091fd94 | 1426 | rcu_read_lock(); |
6ff37972 | 1427 | sem_lock_and_putref(sma); |
1da177e4 | 1428 | if (sma->sem_perm.deleted) { |
1da177e4 | 1429 | err = -EIDRM; |
873be93b | 1430 | goto out_unlock; |
1da177e4 LT |
1431 | } |
1432 | ||
1433 | for (i = 0; i < nsems; i++) | |
1434 | sma->sem_base[i].semval = sem_io[i]; | |
4daa28f6 | 1435 | |
115d40db | 1436 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 | 1437 | list_for_each_entry(un, &sma->list_id, list_id) { |
1da177e4 LT |
1438 | for (i = 0; i < nsems; i++) |
1439 | un->semadj[i] = 0; | |
4daa28f6 | 1440 | } |
1da177e4 LT |
1441 | sma->sem_ctime = get_seconds(); |
1442 | /* maybe some queued-up processes were waiting for this */ | |
0a2b9d4c | 1443 | do_smart_update(sma, NULL, 0, 0, &tasks); |
1da177e4 LT |
1444 | err = 0; |
1445 | goto out_unlock; | |
1446 | } | |
e1fd1f49 | 1447 | /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */ |
1da177e4 LT |
1448 | } |
1449 | err = -EINVAL; | |
c728b9c8 LT |
1450 | if (semnum < 0 || semnum >= nsems) |
1451 | goto out_rcu_wakeup; | |
1da177e4 | 1452 | |
6062a8dc | 1453 | sem_lock(sma, NULL, -1); |
873be93b MS |
1454 | if (sma->sem_perm.deleted) { |
1455 | err = -EIDRM; | |
1456 | goto out_unlock; | |
1457 | } | |
1da177e4 LT |
1458 | curr = &sma->sem_base[semnum]; |
1459 | ||
1460 | switch (cmd) { | |
1461 | case GETVAL: | |
1462 | err = curr->semval; | |
1463 | goto out_unlock; | |
1464 | case GETPID: | |
1465 | err = curr->sempid; | |
1466 | goto out_unlock; | |
1467 | case GETNCNT: | |
1468 | err = count_semncnt(sma,semnum); | |
1469 | goto out_unlock; | |
1470 | case GETZCNT: | |
1471 | err = count_semzcnt(sma,semnum); | |
1472 | goto out_unlock; | |
1da177e4 | 1473 | } |
16df3674 | 1474 | |
1da177e4 | 1475 | out_unlock: |
6062a8dc | 1476 | sem_unlock(sma, -1); |
c728b9c8 | 1477 | out_rcu_wakeup: |
6d49dab8 | 1478 | rcu_read_unlock(); |
0a2b9d4c | 1479 | wake_up_sem_queue_do(&tasks); |
1da177e4 LT |
1480 | out_free: |
1481 | if(sem_io != fast_sem_io) | |
1482 | ipc_free(sem_io, sizeof(ushort)*nsems); | |
1483 | return err; | |
1484 | } | |
1485 | ||
016d7132 PP |
1486 | static inline unsigned long |
1487 | copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version) | |
1da177e4 LT |
1488 | { |
1489 | switch(version) { | |
1490 | case IPC_64: | |
016d7132 | 1491 | if (copy_from_user(out, buf, sizeof(*out))) |
1da177e4 | 1492 | return -EFAULT; |
1da177e4 | 1493 | return 0; |
1da177e4 LT |
1494 | case IPC_OLD: |
1495 | { | |
1496 | struct semid_ds tbuf_old; | |
1497 | ||
1498 | if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old))) | |
1499 | return -EFAULT; | |
1500 | ||
016d7132 PP |
1501 | out->sem_perm.uid = tbuf_old.sem_perm.uid; |
1502 | out->sem_perm.gid = tbuf_old.sem_perm.gid; | |
1503 | out->sem_perm.mode = tbuf_old.sem_perm.mode; | |
1da177e4 LT |
1504 | |
1505 | return 0; | |
1506 | } | |
1507 | default: | |
1508 | return -EINVAL; | |
1509 | } | |
1510 | } | |
1511 | ||
522bb2a2 | 1512 | /* |
33b74669 | 1513 | * This function handles some semctl commands which require the rwsem |
522bb2a2 | 1514 | * to be held in write mode. |
33b74669 | 1515 | * NOTE: no locks must be held, the rwsem is taken inside this function. |
522bb2a2 | 1516 | */ |
21a4826a | 1517 | static int semctl_down(struct ipc_namespace *ns, int semid, |
e1fd1f49 | 1518 | int cmd, int version, void __user *p) |
1da177e4 LT |
1519 | { |
1520 | struct sem_array *sma; | |
1521 | int err; | |
016d7132 | 1522 | struct semid64_ds semid64; |
1da177e4 LT |
1523 | struct kern_ipc_perm *ipcp; |
1524 | ||
1525 | if(cmd == IPC_SET) { | |
e1fd1f49 | 1526 | if (copy_semid_from_user(&semid64, p, version)) |
1da177e4 | 1527 | return -EFAULT; |
1da177e4 | 1528 | } |
073115d6 | 1529 | |
33b74669 | 1530 | down_write(&sem_ids(ns).rwsem); |
ac9bc6e3 DB |
1531 | rcu_read_lock(); |
1532 | ||
16df3674 DB |
1533 | ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd, |
1534 | &semid64.sem_perm, 0); | |
ac9bc6e3 DB |
1535 | if (IS_ERR(ipcp)) { |
1536 | err = PTR_ERR(ipcp); | |
ac9bc6e3 DB |
1537 | goto out_unlock1; |
1538 | } | |
073115d6 | 1539 | |
a5f75e7f | 1540 | sma = container_of(ipcp, struct sem_array, sem_perm); |
1da177e4 LT |
1541 | |
1542 | err = security_sem_semctl(sma, cmd); | |
ac9bc6e3 DB |
1543 | if (err) |
1544 | goto out_unlock1; | |
1da177e4 | 1545 | |
ac9bc6e3 | 1546 | switch (cmd) { |
1da177e4 | 1547 | case IPC_RMID: |
6062a8dc | 1548 | sem_lock(sma, NULL, -1); |
ac9bc6e3 | 1549 | /* freeary unlocks the ipc object and rcu */ |
01b8b07a | 1550 | freeary(ns, ipcp); |
522bb2a2 | 1551 | goto out_up; |
1da177e4 | 1552 | case IPC_SET: |
6062a8dc | 1553 | sem_lock(sma, NULL, -1); |
1efdb69b EB |
1554 | err = ipc_update_perm(&semid64.sem_perm, ipcp); |
1555 | if (err) | |
ac9bc6e3 | 1556 | goto out_unlock0; |
1da177e4 | 1557 | sma->sem_ctime = get_seconds(); |
1da177e4 LT |
1558 | break; |
1559 | default: | |
1da177e4 | 1560 | err = -EINVAL; |
ac9bc6e3 | 1561 | goto out_unlock1; |
1da177e4 | 1562 | } |
1da177e4 | 1563 | |
ac9bc6e3 | 1564 | out_unlock0: |
6062a8dc | 1565 | sem_unlock(sma, -1); |
ac9bc6e3 | 1566 | out_unlock1: |
6d49dab8 | 1567 | rcu_read_unlock(); |
522bb2a2 | 1568 | out_up: |
33b74669 | 1569 | up_write(&sem_ids(ns).rwsem); |
1da177e4 LT |
1570 | return err; |
1571 | } | |
1572 | ||
e1fd1f49 | 1573 | SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg) |
1da177e4 | 1574 | { |
1da177e4 | 1575 | int version; |
e3893534 | 1576 | struct ipc_namespace *ns; |
e1fd1f49 | 1577 | void __user *p = (void __user *)arg; |
1da177e4 LT |
1578 | |
1579 | if (semid < 0) | |
1580 | return -EINVAL; | |
1581 | ||
1582 | version = ipc_parse_version(&cmd); | |
e3893534 | 1583 | ns = current->nsproxy->ipc_ns; |
1da177e4 LT |
1584 | |
1585 | switch(cmd) { | |
1586 | case IPC_INFO: | |
1587 | case SEM_INFO: | |
4b9fcb0e | 1588 | case IPC_STAT: |
1da177e4 | 1589 | case SEM_STAT: |
e1fd1f49 | 1590 | return semctl_nolock(ns, semid, cmd, version, p); |
1da177e4 LT |
1591 | case GETALL: |
1592 | case GETVAL: | |
1593 | case GETPID: | |
1594 | case GETNCNT: | |
1595 | case GETZCNT: | |
1da177e4 | 1596 | case SETALL: |
e1fd1f49 AV |
1597 | return semctl_main(ns, semid, semnum, cmd, p); |
1598 | case SETVAL: | |
1599 | return semctl_setval(ns, semid, semnum, arg); | |
1da177e4 LT |
1600 | case IPC_RMID: |
1601 | case IPC_SET: | |
e1fd1f49 | 1602 | return semctl_down(ns, semid, cmd, version, p); |
1da177e4 LT |
1603 | default: |
1604 | return -EINVAL; | |
1605 | } | |
1606 | } | |
1607 | ||
1da177e4 LT |
1608 | /* If the task doesn't already have a undo_list, then allocate one |
1609 | * here. We guarantee there is only one thread using this undo list, | |
1610 | * and current is THE ONE | |
1611 | * | |
1612 | * If this allocation and assignment succeeds, but later | |
1613 | * portions of this code fail, there is no need to free the sem_undo_list. | |
1614 | * Just let it stay associated with the task, and it'll be freed later | |
1615 | * at exit time. | |
1616 | * | |
1617 | * This can block, so callers must hold no locks. | |
1618 | */ | |
1619 | static inline int get_undo_list(struct sem_undo_list **undo_listp) | |
1620 | { | |
1621 | struct sem_undo_list *undo_list; | |
1da177e4 LT |
1622 | |
1623 | undo_list = current->sysvsem.undo_list; | |
1624 | if (!undo_list) { | |
2453a306 | 1625 | undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL); |
1da177e4 LT |
1626 | if (undo_list == NULL) |
1627 | return -ENOMEM; | |
00a5dfdb | 1628 | spin_lock_init(&undo_list->lock); |
1da177e4 | 1629 | atomic_set(&undo_list->refcnt, 1); |
4daa28f6 MS |
1630 | INIT_LIST_HEAD(&undo_list->list_proc); |
1631 | ||
1da177e4 LT |
1632 | current->sysvsem.undo_list = undo_list; |
1633 | } | |
1634 | *undo_listp = undo_list; | |
1635 | return 0; | |
1636 | } | |
1637 | ||
bf17bb71 | 1638 | static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid) |
1da177e4 | 1639 | { |
bf17bb71 | 1640 | struct sem_undo *un; |
4daa28f6 | 1641 | |
bf17bb71 NP |
1642 | list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) { |
1643 | if (un->semid == semid) | |
1644 | return un; | |
1da177e4 | 1645 | } |
4daa28f6 | 1646 | return NULL; |
1da177e4 LT |
1647 | } |
1648 | ||
bf17bb71 NP |
1649 | static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid) |
1650 | { | |
1651 | struct sem_undo *un; | |
1652 | ||
1653 | assert_spin_locked(&ulp->lock); | |
1654 | ||
1655 | un = __lookup_undo(ulp, semid); | |
1656 | if (un) { | |
1657 | list_del_rcu(&un->list_proc); | |
1658 | list_add_rcu(&un->list_proc, &ulp->list_proc); | |
1659 | } | |
1660 | return un; | |
1661 | } | |
1662 | ||
4daa28f6 MS |
1663 | /** |
1664 | * find_alloc_undo - Lookup (and if not present create) undo array | |
1665 | * @ns: namespace | |
1666 | * @semid: semaphore array id | |
1667 | * | |
1668 | * The function looks up (and if not present creates) the undo structure. | |
1669 | * The size of the undo structure depends on the size of the semaphore | |
1670 | * array, thus the alloc path is not that straightforward. | |
380af1b3 MS |
1671 | * Lifetime-rules: sem_undo is rcu-protected, on success, the function |
1672 | * performs a rcu_read_lock(). | |
4daa28f6 MS |
1673 | */ |
1674 | static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid) | |
1da177e4 LT |
1675 | { |
1676 | struct sem_array *sma; | |
1677 | struct sem_undo_list *ulp; | |
1678 | struct sem_undo *un, *new; | |
6062a8dc | 1679 | int nsems, error; |
1da177e4 LT |
1680 | |
1681 | error = get_undo_list(&ulp); | |
1682 | if (error) | |
1683 | return ERR_PTR(error); | |
1684 | ||
380af1b3 | 1685 | rcu_read_lock(); |
c530c6ac | 1686 | spin_lock(&ulp->lock); |
1da177e4 | 1687 | un = lookup_undo(ulp, semid); |
c530c6ac | 1688 | spin_unlock(&ulp->lock); |
1da177e4 LT |
1689 | if (likely(un!=NULL)) |
1690 | goto out; | |
1691 | ||
1692 | /* no undo structure around - allocate one. */ | |
4daa28f6 | 1693 | /* step 1: figure out the size of the semaphore array */ |
16df3674 DB |
1694 | sma = sem_obtain_object_check(ns, semid); |
1695 | if (IS_ERR(sma)) { | |
1696 | rcu_read_unlock(); | |
4de85cd6 | 1697 | return ERR_CAST(sma); |
16df3674 | 1698 | } |
023a5355 | 1699 | |
1da177e4 | 1700 | nsems = sma->sem_nsems; |
6062a8dc RR |
1701 | if (!ipc_rcu_getref(sma)) { |
1702 | rcu_read_unlock(); | |
1703 | un = ERR_PTR(-EIDRM); | |
1704 | goto out; | |
1705 | } | |
16df3674 | 1706 | rcu_read_unlock(); |
1da177e4 | 1707 | |
4daa28f6 | 1708 | /* step 2: allocate new undo structure */ |
4668edc3 | 1709 | new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL); |
1da177e4 | 1710 | if (!new) { |
e84ca333 | 1711 | ipc_rcu_putref(sma, ipc_rcu_free); |
1da177e4 LT |
1712 | return ERR_PTR(-ENOMEM); |
1713 | } | |
1da177e4 | 1714 | |
380af1b3 | 1715 | /* step 3: Acquire the lock on semaphore array */ |
4091fd94 | 1716 | rcu_read_lock(); |
6ff37972 | 1717 | sem_lock_and_putref(sma); |
1da177e4 | 1718 | if (sma->sem_perm.deleted) { |
6062a8dc | 1719 | sem_unlock(sma, -1); |
6d49dab8 | 1720 | rcu_read_unlock(); |
1da177e4 LT |
1721 | kfree(new); |
1722 | un = ERR_PTR(-EIDRM); | |
1723 | goto out; | |
1724 | } | |
380af1b3 MS |
1725 | spin_lock(&ulp->lock); |
1726 | ||
1727 | /* | |
1728 | * step 4: check for races: did someone else allocate the undo struct? | |
1729 | */ | |
1730 | un = lookup_undo(ulp, semid); | |
1731 | if (un) { | |
1732 | kfree(new); | |
1733 | goto success; | |
1734 | } | |
4daa28f6 MS |
1735 | /* step 5: initialize & link new undo structure */ |
1736 | new->semadj = (short *) &new[1]; | |
380af1b3 | 1737 | new->ulp = ulp; |
4daa28f6 MS |
1738 | new->semid = semid; |
1739 | assert_spin_locked(&ulp->lock); | |
380af1b3 | 1740 | list_add_rcu(&new->list_proc, &ulp->list_proc); |
115d40db | 1741 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 | 1742 | list_add(&new->list_id, &sma->list_id); |
380af1b3 | 1743 | un = new; |
4daa28f6 | 1744 | |
380af1b3 | 1745 | success: |
c530c6ac | 1746 | spin_unlock(&ulp->lock); |
6062a8dc | 1747 | sem_unlock(sma, -1); |
1da177e4 LT |
1748 | out: |
1749 | return un; | |
1750 | } | |
1751 | ||
c61284e9 MS |
1752 | |
1753 | /** | |
1754 | * get_queue_result - Retrieve the result code from sem_queue | |
1755 | * @q: Pointer to queue structure | |
1756 | * | |
1757 | * Retrieve the return code from the pending queue. If IN_WAKEUP is found in | |
1758 | * q->status, then we must loop until the value is replaced with the final | |
1759 | * value: This may happen if a task is woken up by an unrelated event (e.g. | |
1760 | * signal) and in parallel the task is woken up by another task because it got | |
1761 | * the requested semaphores. | |
1762 | * | |
1763 | * The function can be called with or without holding the semaphore spinlock. | |
1764 | */ | |
1765 | static int get_queue_result(struct sem_queue *q) | |
1766 | { | |
1767 | int error; | |
1768 | ||
1769 | error = q->status; | |
1770 | while (unlikely(error == IN_WAKEUP)) { | |
1771 | cpu_relax(); | |
1772 | error = q->status; | |
1773 | } | |
1774 | ||
1775 | return error; | |
1776 | } | |
1777 | ||
d5460c99 HC |
1778 | SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops, |
1779 | unsigned, nsops, const struct timespec __user *, timeout) | |
1da177e4 LT |
1780 | { |
1781 | int error = -EINVAL; | |
1782 | struct sem_array *sma; | |
1783 | struct sembuf fast_sops[SEMOPM_FAST]; | |
1784 | struct sembuf* sops = fast_sops, *sop; | |
1785 | struct sem_undo *un; | |
6062a8dc | 1786 | int undos = 0, alter = 0, max, locknum; |
1da177e4 LT |
1787 | struct sem_queue queue; |
1788 | unsigned long jiffies_left = 0; | |
e3893534 | 1789 | struct ipc_namespace *ns; |
0a2b9d4c | 1790 | struct list_head tasks; |
e3893534 KK |
1791 | |
1792 | ns = current->nsproxy->ipc_ns; | |
1da177e4 LT |
1793 | |
1794 | if (nsops < 1 || semid < 0) | |
1795 | return -EINVAL; | |
e3893534 | 1796 | if (nsops > ns->sc_semopm) |
1da177e4 LT |
1797 | return -E2BIG; |
1798 | if(nsops > SEMOPM_FAST) { | |
1799 | sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL); | |
1800 | if(sops==NULL) | |
1801 | return -ENOMEM; | |
1802 | } | |
1803 | if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) { | |
1804 | error=-EFAULT; | |
1805 | goto out_free; | |
1806 | } | |
1807 | if (timeout) { | |
1808 | struct timespec _timeout; | |
1809 | if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) { | |
1810 | error = -EFAULT; | |
1811 | goto out_free; | |
1812 | } | |
1813 | if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 || | |
1814 | _timeout.tv_nsec >= 1000000000L) { | |
1815 | error = -EINVAL; | |
1816 | goto out_free; | |
1817 | } | |
1818 | jiffies_left = timespec_to_jiffies(&_timeout); | |
1819 | } | |
1820 | max = 0; | |
1821 | for (sop = sops; sop < sops + nsops; sop++) { | |
1822 | if (sop->sem_num >= max) | |
1823 | max = sop->sem_num; | |
1824 | if (sop->sem_flg & SEM_UNDO) | |
b78755ab MS |
1825 | undos = 1; |
1826 | if (sop->sem_op != 0) | |
1da177e4 LT |
1827 | alter = 1; |
1828 | } | |
1da177e4 | 1829 | |
6062a8dc RR |
1830 | INIT_LIST_HEAD(&tasks); |
1831 | ||
1da177e4 | 1832 | if (undos) { |
6062a8dc | 1833 | /* On success, find_alloc_undo takes the rcu_read_lock */ |
4daa28f6 | 1834 | un = find_alloc_undo(ns, semid); |
1da177e4 LT |
1835 | if (IS_ERR(un)) { |
1836 | error = PTR_ERR(un); | |
1837 | goto out_free; | |
1838 | } | |
6062a8dc | 1839 | } else { |
1da177e4 | 1840 | un = NULL; |
6062a8dc RR |
1841 | rcu_read_lock(); |
1842 | } | |
1da177e4 | 1843 | |
16df3674 | 1844 | sma = sem_obtain_object_check(ns, semid); |
023a5355 | 1845 | if (IS_ERR(sma)) { |
6062a8dc | 1846 | rcu_read_unlock(); |
023a5355 | 1847 | error = PTR_ERR(sma); |
1da177e4 | 1848 | goto out_free; |
023a5355 ND |
1849 | } |
1850 | ||
16df3674 | 1851 | error = -EFBIG; |
c728b9c8 LT |
1852 | if (max >= sma->sem_nsems) |
1853 | goto out_rcu_wakeup; | |
16df3674 DB |
1854 | |
1855 | error = -EACCES; | |
c728b9c8 LT |
1856 | if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) |
1857 | goto out_rcu_wakeup; | |
16df3674 DB |
1858 | |
1859 | error = security_sem_semop(sma, sops, nsops, alter); | |
c728b9c8 LT |
1860 | if (error) |
1861 | goto out_rcu_wakeup; | |
16df3674 | 1862 | |
873be93b MS |
1863 | error = -EIDRM; |
1864 | locknum = sem_lock(sma, sops, nsops); | |
1865 | if (sma->sem_perm.deleted) | |
1866 | goto out_unlock_free; | |
1da177e4 | 1867 | /* |
4daa28f6 | 1868 | * semid identifiers are not unique - find_alloc_undo may have |
1da177e4 | 1869 | * allocated an undo structure, it was invalidated by an RMID |
4daa28f6 | 1870 | * and now a new array with received the same id. Check and fail. |
25985edc | 1871 | * This case can be detected checking un->semid. The existence of |
380af1b3 | 1872 | * "un" itself is guaranteed by rcu. |
1da177e4 | 1873 | */ |
6062a8dc RR |
1874 | if (un && un->semid == -1) |
1875 | goto out_unlock_free; | |
4daa28f6 | 1876 | |
b56e88e2 MS |
1877 | error = perform_atomic_semop(sma, sops, nsops, un, |
1878 | task_tgid_vnr(current)); | |
e556ea01 MS |
1879 | if (error == 0) { |
1880 | /* If the operation was successful, then do | |
1881 | * the required updates. | |
1882 | */ | |
1883 | if (alter) | |
0a2b9d4c | 1884 | do_smart_update(sma, sops, nsops, 1, &tasks); |
e556ea01 MS |
1885 | else |
1886 | set_semotime(sma, sops); | |
1da177e4 | 1887 | } |
e556ea01 MS |
1888 | if (error <= 0) |
1889 | goto out_unlock_free; | |
1da177e4 LT |
1890 | |
1891 | /* We need to sleep on this operation, so we put the current | |
1892 | * task into the pending queue and go to sleep. | |
1893 | */ | |
1894 | ||
1da177e4 LT |
1895 | queue.sops = sops; |
1896 | queue.nsops = nsops; | |
1897 | queue.undo = un; | |
b488893a | 1898 | queue.pid = task_tgid_vnr(current); |
1da177e4 | 1899 | queue.alter = alter; |
1da177e4 | 1900 | |
b97e820f MS |
1901 | if (nsops == 1) { |
1902 | struct sem *curr; | |
1903 | curr = &sma->sem_base[sops->sem_num]; | |
1904 | ||
e5639c52 MS |
1905 | if (alter) { |
1906 | if (sma->complex_count) { | |
1907 | list_add_tail(&queue.list, | |
1908 | &sma->pending_alter); | |
1909 | } else { | |
1910 | ||
1911 | list_add_tail(&queue.list, | |
1912 | &curr->pending_alter); | |
1913 | } | |
1914 | } else { | |
ab63bc97 | 1915 | list_add_tail(&queue.list, &curr->pending_const); |
e5639c52 | 1916 | } |
b97e820f | 1917 | } else { |
e5639c52 MS |
1918 | if (!sma->complex_count) |
1919 | merge_queues(sma); | |
1920 | ||
9f1bc2c9 | 1921 | if (alter) |
ab63bc97 | 1922 | list_add_tail(&queue.list, &sma->pending_alter); |
9f1bc2c9 | 1923 | else |
ab63bc97 MS |
1924 | list_add_tail(&queue.list, &sma->pending_const); |
1925 | ||
b97e820f MS |
1926 | sma->complex_count++; |
1927 | } | |
1928 | ||
1da177e4 LT |
1929 | queue.status = -EINTR; |
1930 | queue.sleeper = current; | |
0b0577f6 MS |
1931 | |
1932 | sleep_again: | |
1da177e4 | 1933 | current->state = TASK_INTERRUPTIBLE; |
6062a8dc | 1934 | sem_unlock(sma, locknum); |
6d49dab8 | 1935 | rcu_read_unlock(); |
1da177e4 LT |
1936 | |
1937 | if (timeout) | |
1938 | jiffies_left = schedule_timeout(jiffies_left); | |
1939 | else | |
1940 | schedule(); | |
1941 | ||
c61284e9 | 1942 | error = get_queue_result(&queue); |
1da177e4 LT |
1943 | |
1944 | if (error != -EINTR) { | |
1945 | /* fast path: update_queue already obtained all requested | |
c61284e9 MS |
1946 | * resources. |
1947 | * Perform a smp_mb(): User space could assume that semop() | |
1948 | * is a memory barrier: Without the mb(), the cpu could | |
1949 | * speculatively read in user space stale data that was | |
1950 | * overwritten by the previous owner of the semaphore. | |
1951 | */ | |
1952 | smp_mb(); | |
1953 | ||
1da177e4 LT |
1954 | goto out_free; |
1955 | } | |
1956 | ||
321310ce | 1957 | rcu_read_lock(); |
6062a8dc | 1958 | sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum); |
d694ad62 MS |
1959 | |
1960 | /* | |
1961 | * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing. | |
1962 | */ | |
1963 | error = get_queue_result(&queue); | |
1964 | ||
1965 | /* | |
1966 | * Array removed? If yes, leave without sem_unlock(). | |
1967 | */ | |
023a5355 | 1968 | if (IS_ERR(sma)) { |
321310ce | 1969 | rcu_read_unlock(); |
1da177e4 LT |
1970 | goto out_free; |
1971 | } | |
1972 | ||
c61284e9 | 1973 | |
1da177e4 | 1974 | /* |
d694ad62 MS |
1975 | * If queue.status != -EINTR we are woken up by another process. |
1976 | * Leave without unlink_queue(), but with sem_unlock(). | |
1da177e4 | 1977 | */ |
c61284e9 | 1978 | |
1da177e4 LT |
1979 | if (error != -EINTR) { |
1980 | goto out_unlock_free; | |
1981 | } | |
1982 | ||
1983 | /* | |
1984 | * If an interrupt occurred we have to clean up the queue | |
1985 | */ | |
1986 | if (timeout && jiffies_left == 0) | |
1987 | error = -EAGAIN; | |
0b0577f6 MS |
1988 | |
1989 | /* | |
1990 | * If the wakeup was spurious, just retry | |
1991 | */ | |
1992 | if (error == -EINTR && !signal_pending(current)) | |
1993 | goto sleep_again; | |
1994 | ||
b97e820f | 1995 | unlink_queue(sma, &queue); |
1da177e4 LT |
1996 | |
1997 | out_unlock_free: | |
6062a8dc | 1998 | sem_unlock(sma, locknum); |
c728b9c8 | 1999 | out_rcu_wakeup: |
6d49dab8 | 2000 | rcu_read_unlock(); |
0a2b9d4c | 2001 | wake_up_sem_queue_do(&tasks); |
1da177e4 LT |
2002 | out_free: |
2003 | if(sops != fast_sops) | |
2004 | kfree(sops); | |
2005 | return error; | |
2006 | } | |
2007 | ||
d5460c99 HC |
2008 | SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops, |
2009 | unsigned, nsops) | |
1da177e4 LT |
2010 | { |
2011 | return sys_semtimedop(semid, tsops, nsops, NULL); | |
2012 | } | |
2013 | ||
2014 | /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between | |
2015 | * parent and child tasks. | |
1da177e4 LT |
2016 | */ |
2017 | ||
2018 | int copy_semundo(unsigned long clone_flags, struct task_struct *tsk) | |
2019 | { | |
2020 | struct sem_undo_list *undo_list; | |
2021 | int error; | |
2022 | ||
2023 | if (clone_flags & CLONE_SYSVSEM) { | |
2024 | error = get_undo_list(&undo_list); | |
2025 | if (error) | |
2026 | return error; | |
1da177e4 LT |
2027 | atomic_inc(&undo_list->refcnt); |
2028 | tsk->sysvsem.undo_list = undo_list; | |
2029 | } else | |
2030 | tsk->sysvsem.undo_list = NULL; | |
2031 | ||
2032 | return 0; | |
2033 | } | |
2034 | ||
2035 | /* | |
2036 | * add semadj values to semaphores, free undo structures. | |
2037 | * undo structures are not freed when semaphore arrays are destroyed | |
2038 | * so some of them may be out of date. | |
2039 | * IMPLEMENTATION NOTE: There is some confusion over whether the | |
2040 | * set of adjustments that needs to be done should be done in an atomic | |
2041 | * manner or not. That is, if we are attempting to decrement the semval | |
2042 | * should we queue up and wait until we can do so legally? | |
2043 | * The original implementation attempted to do this (queue and wait). | |
2044 | * The current implementation does not do so. The POSIX standard | |
2045 | * and SVID should be consulted to determine what behavior is mandated. | |
2046 | */ | |
2047 | void exit_sem(struct task_struct *tsk) | |
2048 | { | |
4daa28f6 | 2049 | struct sem_undo_list *ulp; |
1da177e4 | 2050 | |
4daa28f6 MS |
2051 | ulp = tsk->sysvsem.undo_list; |
2052 | if (!ulp) | |
1da177e4 | 2053 | return; |
9edff4ab | 2054 | tsk->sysvsem.undo_list = NULL; |
1da177e4 | 2055 | |
4daa28f6 | 2056 | if (!atomic_dec_and_test(&ulp->refcnt)) |
1da177e4 LT |
2057 | return; |
2058 | ||
380af1b3 | 2059 | for (;;) { |
1da177e4 | 2060 | struct sem_array *sma; |
380af1b3 | 2061 | struct sem_undo *un; |
0a2b9d4c | 2062 | struct list_head tasks; |
6062a8dc | 2063 | int semid, i; |
4daa28f6 | 2064 | |
380af1b3 | 2065 | rcu_read_lock(); |
05725f7e JP |
2066 | un = list_entry_rcu(ulp->list_proc.next, |
2067 | struct sem_undo, list_proc); | |
04d2af28 HK |
2068 | if (&un->list_proc == &ulp->list_proc) { |
2069 | /* | |
2070 | * We must wait for freeary() before freeing this ulp, | |
2071 | * in case we raced with last sem_undo. There is a small | |
2072 | * possibility where we exit while freeary() didn't | |
2073 | * finish unlocking sem_undo_list. | |
2074 | */ | |
2075 | spin_unlock_wait(&ulp->lock); | |
2076 | rcu_read_unlock(); | |
2077 | break; | |
2078 | } | |
2079 | spin_lock(&ulp->lock); | |
2080 | semid = un->semid; | |
2081 | spin_unlock(&ulp->lock); | |
4daa28f6 | 2082 | |
04d2af28 | 2083 | /* exit_sem raced with IPC_RMID, nothing to do */ |
6062a8dc RR |
2084 | if (semid == -1) { |
2085 | rcu_read_unlock(); | |
04d2af28 | 2086 | continue; |
6062a8dc | 2087 | } |
1da177e4 | 2088 | |
04d2af28 | 2089 | sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid); |
380af1b3 | 2090 | /* exit_sem raced with IPC_RMID, nothing to do */ |
6062a8dc RR |
2091 | if (IS_ERR(sma)) { |
2092 | rcu_read_unlock(); | |
380af1b3 | 2093 | continue; |
6062a8dc | 2094 | } |
1da177e4 | 2095 | |
6062a8dc | 2096 | sem_lock(sma, NULL, -1); |
873be93b MS |
2097 | /* exit_sem raced with IPC_RMID, nothing to do */ |
2098 | if (sma->sem_perm.deleted) { | |
2099 | sem_unlock(sma, -1); | |
2100 | rcu_read_unlock(); | |
2101 | continue; | |
2102 | } | |
bf17bb71 | 2103 | un = __lookup_undo(ulp, semid); |
380af1b3 MS |
2104 | if (un == NULL) { |
2105 | /* exit_sem raced with IPC_RMID+semget() that created | |
2106 | * exactly the same semid. Nothing to do. | |
2107 | */ | |
6062a8dc | 2108 | sem_unlock(sma, -1); |
6d49dab8 | 2109 | rcu_read_unlock(); |
380af1b3 MS |
2110 | continue; |
2111 | } | |
2112 | ||
2113 | /* remove un from the linked lists */ | |
115d40db | 2114 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 MS |
2115 | list_del(&un->list_id); |
2116 | ||
380af1b3 MS |
2117 | spin_lock(&ulp->lock); |
2118 | list_del_rcu(&un->list_proc); | |
2119 | spin_unlock(&ulp->lock); | |
2120 | ||
4daa28f6 MS |
2121 | /* perform adjustments registered in un */ |
2122 | for (i = 0; i < sma->sem_nsems; i++) { | |
5f921ae9 | 2123 | struct sem * semaphore = &sma->sem_base[i]; |
4daa28f6 MS |
2124 | if (un->semadj[i]) { |
2125 | semaphore->semval += un->semadj[i]; | |
1da177e4 LT |
2126 | /* |
2127 | * Range checks of the new semaphore value, | |
2128 | * not defined by sus: | |
2129 | * - Some unices ignore the undo entirely | |
2130 | * (e.g. HP UX 11i 11.22, Tru64 V5.1) | |
2131 | * - some cap the value (e.g. FreeBSD caps | |
2132 | * at 0, but doesn't enforce SEMVMX) | |
2133 | * | |
2134 | * Linux caps the semaphore value, both at 0 | |
2135 | * and at SEMVMX. | |
2136 | * | |
2137 | * Manfred <manfred@colorfullife.com> | |
2138 | */ | |
5f921ae9 IM |
2139 | if (semaphore->semval < 0) |
2140 | semaphore->semval = 0; | |
2141 | if (semaphore->semval > SEMVMX) | |
2142 | semaphore->semval = SEMVMX; | |
b488893a | 2143 | semaphore->sempid = task_tgid_vnr(current); |
1da177e4 LT |
2144 | } |
2145 | } | |
1da177e4 | 2146 | /* maybe some queued-up processes were waiting for this */ |
0a2b9d4c MS |
2147 | INIT_LIST_HEAD(&tasks); |
2148 | do_smart_update(sma, NULL, 0, 1, &tasks); | |
6062a8dc | 2149 | sem_unlock(sma, -1); |
6d49dab8 | 2150 | rcu_read_unlock(); |
0a2b9d4c | 2151 | wake_up_sem_queue_do(&tasks); |
380af1b3 | 2152 | |
693a8b6e | 2153 | kfree_rcu(un, rcu); |
1da177e4 | 2154 | } |
4daa28f6 | 2155 | kfree(ulp); |
1da177e4 LT |
2156 | } |
2157 | ||
2158 | #ifdef CONFIG_PROC_FS | |
19b4946c | 2159 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it) |
1da177e4 | 2160 | { |
1efdb69b | 2161 | struct user_namespace *user_ns = seq_user_ns(s); |
19b4946c | 2162 | struct sem_array *sma = it; |
bf6830ad MS |
2163 | time_t sem_otime; |
2164 | ||
83aeb6e3 MS |
2165 | /* |
2166 | * The proc interface isn't aware of sem_lock(), it calls | |
2167 | * ipc_lock_object() directly (in sysvipc_find_ipc). | |
2168 | * In order to stay compatible with sem_lock(), we must wait until | |
2169 | * all simple semop() calls have left their critical regions. | |
2170 | */ | |
2171 | sem_wait_array(sma); | |
2172 | ||
bf6830ad | 2173 | sem_otime = get_semotime(sma); |
19b4946c MW |
2174 | |
2175 | return seq_printf(s, | |
b97e820f | 2176 | "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n", |
19b4946c | 2177 | sma->sem_perm.key, |
7ca7e564 | 2178 | sma->sem_perm.id, |
19b4946c MW |
2179 | sma->sem_perm.mode, |
2180 | sma->sem_nsems, | |
1efdb69b EB |
2181 | from_kuid_munged(user_ns, sma->sem_perm.uid), |
2182 | from_kgid_munged(user_ns, sma->sem_perm.gid), | |
2183 | from_kuid_munged(user_ns, sma->sem_perm.cuid), | |
2184 | from_kgid_munged(user_ns, sma->sem_perm.cgid), | |
bf6830ad | 2185 | sem_otime, |
19b4946c | 2186 | sma->sem_ctime); |
1da177e4 LT |
2187 | } |
2188 | #endif |